Fluid mechanics of flow metering

Wolfgang Merzkirch (Editor) Fluid Mechanics of Flow Metering Wolfgang Merzkirch (Editor) Fluid Mechanics of Flow Met...

422 downloads 2565 Views 11MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

Wolfgang Merzkirch (Editor) Fluid Mechanics of Flow Metering

Wolfgang Merzkirch (Editor)

Fluid Mechanics of Flow Metering With 150 Figures

Q - Springer

Editor W o l f g a n g Merzkirch

Authors Wolfgang Merzkirch

Ernst von L a v a n t e

Universitat Duisburg-Essen FB ~ a s c h i n e n w e s e ~ Stromungslehre Schutzenbahn 70 451 17 Essen, Germany

Universitat Duisburg-Essen FB Maschinentechnik Stromungsmaschinen Schiitzenbahn 70 451 17 Essen, Germany

Klaus Gersten

Volker Hans

Franz Peters Venkatesa Vasanta R a m

Universitat Duisburg-Essen FB 12 Maschinenwesen Mess- u n d Regelungstechnik Schiitzenbahn 70 451 17 Essen, Germany

Universitat Bochum Fakultat fiir Maschinenbau Institut Thermo- u n d Fluiddynamik Universitatstr. 150 44780 Bochum, Germany

In cooperation with: Kathrin Kalkuhler, Hans-Dieter PapenfuD, Carsten Ruppel, Franka Schneider, Carsten Wildemann, Wei Xiong

Library of Congress Control Number:

ISBN 3-540-22242- 1 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in other ways, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions ofthe German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under German Copyright Law. Springer is a part of Springer Science+Business Media O Springer-Verlag Berlin

Printed in Germany

Heidelberg 2005

The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: Digital data supplied by editors Cover-Design: Design & Production, Heidelberg 6113020 Rw 5 4 3 2 1 0 Printed on acid-free paper

Preface

A ﬂow meter is a measuring device for determining the ﬂow rate of a ﬂuid, normally in a pipe, e.g., the ﬂow of natural gas, or oil, or waste water. With these instruments, an immense amount of money is controlled, and thus, ﬂow metering has an enormous economical signiﬁcance. In an editorial of the journal “Flow Measurement and Instrumentation” (Vol. 1, no. 1, 1989) it was stated that the costs that are controlled by ﬂow meters are estimated as being worldwide of the order of 10,000 billion US dollars (10,000 milliard EURO) per year. One percent uncertainty in the measurement would still represent a considerable value. The majority of the existing literature on ﬂow metering is authored by electrical and electronic engineers. The emphasis of this literature is on the processing of the signals generated by the ﬂuid ﬂow and on the methods of calibration, by which a relationship between the signals and the desired quantity, the volumetric ﬂow rate in the pipe, is established. For a long time, the ﬁeld of ﬂow metering was disregarded by ﬂuid mechanicists, although it poses challenging problems of ﬂuid dynamics. In view of this situation, the authors of this book started a special research programme at Universität Essen and Universität Bochum in 1995 that was aimed at providing a more substantial understanding of “ﬂuid-mechanical fundamentals of ﬂow metering”, hence the ofﬁcial title of the programme for which ﬁnancial support was granted by Deutsche Forschungsgemeinschaft (“Forschergruppe Strömungsmechanische Grundlagen der Durchﬂussmessung”, DFG Az. ME 484/29). This book summarises the results of this research programme. Two problem areas of particular ﬂuid-mechanical interest are emphasised: the generation of the signals by the ﬂuid ﬂow in the pipe, and the measurement of the volumetric ﬂow rate in pipe ﬂow that is not fully developed, a situation that often occurs in practice and where the state of ﬂow deviates from that during calibration of the meter. The discussion focuses on ﬂows at high Reynolds numbers as they are relevant for many industrial processes, e.g., the transport of natural gas. The book starts with a survey of the theory of turbulent pipe ﬂow; deviations from the fully developed state of ﬂow are characterized; the physics of the signal generation are analyzed for a number of known and newly proposed methods of ﬂow metering; and ways of accounting

VI

Preface

for ﬂow disturbances, i.e., deviations from the fully developed state, either by ﬂow conditioning or correcting the reading of a meter, are presented. The authors are grateful for the support received from Deutsche Forschungsgemeinschaft, and they would like to thank the many students and doctoral students who have worked with them in this programme. They express their particular thanks and appreciation to Dr. Carsten Ruppel who brought all the manuscripts into the form required by the publisher.

Essen, April 2004

W. Merzkirch

Contents

1

2

Fully Developed Turbulent Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Two-Layer Structure for Reτ → ∞ . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Wall Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Core Region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Friction Law for Given Volume Flux . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Velocity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.1 General Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6.2 Determination of the Model Constants . . . . . . . . . . . . . . . . . . 1.7 Inﬂuence of Wall Roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7.1 Hydraulically Deﬁned Roughness . . . . . . . . . . . . . . . . . . . . . . 1.7.2 Natural Roughness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7.3 Velocity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7.4 Geometric Roughness Parameters . . . . . . . . . . . . . . . . . . . . . . 1.7.5 Relation Between Hydraulic and Geometric Roughness Parameters . . . . . . . . . . . . . . . . . . . 1.8 Low Reynolds Number Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

17 18 20

Decay of Disturbances in Turbulent Pipe Flow . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Two-Layer Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Core Region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Fully Developed Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Velocity Distribution in the Core Region . . . . . . . . . . . . . . . . . 2.3.2 Friction Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Eddy-Viscosity Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Eigenvalue Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23 23 24 24 25 28 29 29 31 32 34 34

1 1 4 4 5 8 9 9 11 13 13 14 15 16

VIII

Contents

2.4.2 Limiting Solution for = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Numerical Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Comparisons with Experimental Results . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 Axisymmetric Flow without Swirl . . . . . . . . . . . . . . . . . . . . . . 2.5.2 Axisymmetric Flow with Swirl . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.3 Three-Dimensional Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

4

5

36 36 42 42 43 44 46

Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Formulae for the Optimal Characteristic Parameters . . . . . . . . . . . . . . 3.4 Fully Developed Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Minimal Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7 Experimental Determination of the Characteristic Parameters (Minimal Program) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

56 59

Measurement of Velocity and Turbulence Downstream of Flow Conditioners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Nearﬁeld Downstream of the Conditioners . . . . . . . . . . . . . . . 4.3.2 Redevelopment of the Flow in the Far Field . . . . . . . . . . . . . . 4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61 61 62 65 65 69 75 77

Signal Processing of Complex Modulated Ultrasonic Signals . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Cross-Correlation Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Demodulation by Digital Undersampling . . . . . . . . . . . . . . . . . . . . . . . 5.4 Digital Hilbert-Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1 Undersampled Hilbert-Transform . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Software-Based Hilbert-Transform . . . . . . . . . . . . . . . . . . . . . 5.4.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.4 Analog Electronic Hilbert-Transform . . . . . . . . . . . . . . . . . . . 5.5 Phase Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 Phase Demodulation with Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . 5.7 Analog Signal Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79 79 80 81 83 83 84 85 86 88 89 91 93 94

49 49 50 50 52 53 56

Contents

IX

6

Vortex-Shedding Flow Metering Using Ultrasound . . . . . . . . . . . . . . . . 95 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 6.2 Physical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 6.3 Measurement Principle and Test Arrangement . . . . . . . . . . . . . . . . . . 97 6.4 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 6.5 Comparisons of Pressure and Ultrasonic Signals . . . . . . . . . . . . . . . . . 98 6.6 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.6.1 Large Triangular Bluff Body . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.6.2 Small Triangular Bluff Body . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 6.6.3 T-Shaped Bluff Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 6.6.4 Circular Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 6.6.5 Threaded Control Rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 6.6.6 Pressure Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 6.7 Inﬂuence of Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 6.7.1 Single and Double Elbows . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 6.7.2 Pulsation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 6.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

7

Ultrasonic Gas-Flow Measurement Using Correlation Methods . . . . . . 111 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 7.2 Determination of Traveling Time by Cross-Correlation Functions . . 113 7.3 Physical Quantity Measured by Cross-Correlation Functions . . . . . . 114 7.4 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 7.4.1 Measurement Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 7.4.2 Measurement of Disturbed Flow Proﬁles . . . . . . . . . . . . . . . . . 117 7.5 Multipath Arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 7.6 Tomographic Reconstruction of Flow Proﬁle . . . . . . . . . . . . . . . . . . . . 125 7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

8

Ultrasound Cross-Correlation Flow Meter: Analysis by System Theory and Inﬂuence of Turbulence . . . . . . . . . . . . 129 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 8.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 8.3 System-Theoretical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 8.3.1 Interpretation of the Measurement System as a Linear Time-Invariant System . . . . . . . . . . . . . . . . . . . . . . 133 8.3.2 Impulse Response of the Ultrasound System . . . . . . . . . . . . . 135 8.4 Comparison of Theoretical and Experimental Results . . . . . . . . . . . . 141 8.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

X

9

Contents

Effect of Area Changes in Swirling Flow . . . . . . . . . . . . . . . . . . . . . . . . . 149 9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 9.2 Physical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 9.2.1 Kinematic and Dynamic Characteristics . . . . . . . . . . . . . . . . . 151 9.2.2 Scope of the Present Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 9.3 Mathematical Formulation of the Problem . . . . . . . . . . . . . . . . . . . . . . 153 9.3.1 Solution for the Special Case of Homogeneous Axial Velocity and Solid-Body Rotation at Inlet . . . . . . . . . . . . . . . . 155 9.3.2 The General Case of Arbitrary Proﬁles of Axial and Azimuthal Velocity Components at Inlet . . . . . . . . . . . . . 157 9.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 9.4.1 Veriﬁcation of the Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 9.4.2 The Effects of Passage Contraction . . . . . . . . . . . . . . . . . . . . . 161 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

10 Errors of Turbine Meters Due to Swirl . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 10.2 Physical and Mathematical Background . . . . . . . . . . . . . . . . . . . . . . . . 168 10.2.1 Aerodynamic Torque on the Turbine Flow Meter Rotor . . . . 169 10.2.2 Braking Torque on the Turbine Flow Meter Rotor . . . . . . . . . 178 10.3 Experiments and Data Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 10.3.1 Deﬁnition of Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 10.3.2 Extraction of Error Due to Swirl . . . . . . . . . . . . . . . . . . . . . . . . 181 10.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 10.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 11.2 Theory of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 11.3 Numerical Tools: Fluent and ACHIEVE . . . . . . . . . . . . . . . . . . . . . . . . 188 11.4 Three-Dimensional Simulations: Geometry and Setup . . . . . . . . . . . . 188 11.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 11.6 The Pressure Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 11.7 Secondary Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 11.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 12 How to Design a New Flow Meter from Scratch . . . . . . . . . . . . . . . . . . . 201 12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 12.2 Measuring Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 12.2.1 Flow Rate and Drag Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 12.2.2 Drag Force and Load Cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 12.3 Setup and Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

Contents

XI

12.4 Results for Installations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 12.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body . . . 211 13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 13.2 Numerical Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 13.2.1 Low Mach Number Modiﬁcations . . . . . . . . . . . . . . . . . . . . . . 213 13.2.2 Turbulence Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 13.2.3 Veriﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 13.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 13.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 14 Correction of the Reading of a Flow Meter in Pipe Flow Disturbed by Installation Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223 14.2 Characterization of the Disturbed Flow . . . . . . . . . . . . . . . . . . . . . . . . 225 14.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 14.3.1 Flow Facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 14.3.2 Measurement of Wall Shear Stress . . . . . . . . . . . . . . . . . . . . . . 227 14.3.3 Measurement of Discharge Coefﬁcients . . . . . . . . . . . . . . . . . 229 14.4 Relationship Between Error Shift and Flow Disturbance . . . . . . . . . . 231 14.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 14.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 15 How to Correct the Error Shift of an Ultrasonic Flow Meter Downstream of Installations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 15.2 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 15.2.1 Test Rig . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 15.2.2 Probe Flange . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 15.2.3 Meter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 15.2.4 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 15.2.5 Installations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 15.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 15.4 Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 15.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 15.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

1 Fully Developed Turbulent Pipe Flow Klaus Gersten Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

Incompressible pipe ﬂows are fully developed if the velocities are independent of the axial coordinate. The theory of turbulent pipe ﬂows at high Reynolds numbers leads to analytical expressions for the velocity proﬁle and the friction factor, which contain free constants. One of the latter is the well-known Kármán constant. The free constants can be determined from the best available experimental data. Final formulae for the velocity distribution and the friction factor are derived. Further effects on these laws are also considered, such as wall roughness and low Reynolds number effects.

1.1 Basic Equations Fully developed turbulent pipe ﬂows are the basis for most ﬂow meter concepts and ﬂow meters are usually calibrated in fully developed ﬂows. The fully developed turbulent ﬂow of a ﬂuid with constant physical properties (density ρ∗ , viscosity µ∗ ) in a circular pipe with the radius R∗ = d∗ /2 is considered. Here and in the following text, dimensional quantities are marked by an asterisk. To describe this ﬂow, cylindrical coordinates x∗ , r∗ , θ with corresponding velocities u∗ , v ∗ , w∗ are used, see Fig. 1.1. A ﬂow is designated as fully developed if the distribution of the (time averaged) mean velocity u∗ (r∗ ) is independent of x∗ . Since the ﬂow is axisymmetric the continuity equation reads 1 ∂ r∗ v ∗ ∂u∗ + ∗ =0 (1.1) ∂x∗ r ∂r∗ and leads to v ∗ = 0 (as consequence of the impermeability of the wall). Hence, the two momentum equations reduce to (τ ∗ and v ∗ 2 are also independent of x): ∂p∗ 1 d r∗ τ ∗ − ∗ =0, (1.2) ∗ ∗ r dr ∂x

2

1 Fully Developed Turbulent Pipe Flow

R* r*,v* ,w

x*,u*

Fig. 1.1. Geometry and coordinate system.

ρ

∗

∂p∗ d(v ∗ 2 ) v ∗ 2 − w∗ 2 + + ∗ =0, ∗ ∗ dr r ∂r

(1.3)

where overbars denote the time averaged values while primes denote the turbulent ﬂuctuations. The time averaged pressure p∗ (x∗ , r∗ ) is not independent of r ∗ as in the laminar case. Integrating (1.3) yields ∗

p∗ (x∗ , r∗ ) = p∗w − ρ∗ v ∗ 2 + ρ∗

R r∗

v ∗ 2 − w∗ 2 ∗ dr . r∗

(1.4)

The difference between the local pressure in the ﬂow ﬁeld p∗ and the wall pressure p∗w is due to the turbulent ﬂuctuations and rather small. Since the velocity ﬁeld and, hence, v ∗ 2 and w∗ 2 are independent of x∗ , it follows from (1.4) dp∗ ∂p∗ = w∗ (1.5) ∗ ∂x dx and from (1.2) dp∗w 1 d ∗ ∗ (r =0. (1.6) τ ) − r∗ dr∗ dx∗ The consequence of this equation is that the axial pressure gradient dp ∗w /dx∗ is constant. Integrating (1.6) leads to τ∗ = and particularly for r∗ = R∗ to

dp∗w r∗ dx∗ 2

(1.7)

1.1 Basic Equations

τw∗ =

dp∗w R∗ . dx∗ 2

3

(1.8)

Hence, the wall shear stress τw∗ is proportional to the pressure gradient and can easily be determined experimentally by measuring the pressure gradient. It follows from (1.7) and (1.8) τ∗ =

r∗ ∗ τ , R∗ w

(1.9)

i.e., the local shear stress τ ∗ is proportional to the radius r ∗ . If the (total) shear stress τ ∗ is split into a viscous and a turbulent part τ ∗ = τv∗ + τt∗ = ρ∗ ν ∗

du∗ − ρ∗ u∗ v ∗ dr∗

(1.10)

(ν ∗ = µ∗ /ρ∗ = kinematic viscosity) the force balance in axial direction, (1.9), leads to r∗ du∗ (1.11) ρ∗ ν ∗ ∗ + τt∗ = ∗ τw∗ . dr R Note 1.1 (Sign of Shear Stress) The sign of the shear stress depends on the coordinate system chosen. The shear stress in the fully developed circular pipe ﬂow is negative if cylindrical coordinates are used, cf. Fig. 1.1. In this case, the velocity component v ∗ is positive in radially outward direction.

Equation (1.11) will be applied to solve the following problem: the pressure gradient dp∗w /dx∗ (and hence the wall shear stress τw∗ according to (1.8)) and the quantities R∗ , ρ∗ and ν ∗ are prescribed and the distribution of the velocity gradient ∗ du∗ ∗ ∗ ∗ τw = f r , R , ν , (1.12) dr∗ ρ∗ has to be found. Integration leads ﬁnally to the velocity distribution u ∗ (r∗ ) and to the volume ﬂux. Using the dimensionless quantities τt∗ R∗ u∗τ u∗ −τw∗ r∗ + ∗ + , Reτ = (1.13) r = ∗ , u = ∗ , τt = ∗ , uτ = ∗ R uτ ρ ν∗ τw reduces (1.11) to −

1 du+ + τt+ = r . Reτ dr

(1.14)

Now the function du+ /dr = F (r, Reτ ) which also follows from (1.12) by dimensional analysis must be found. This function is here of particular interest for high Reynolds numbers Reτ . It should be mentioned that (1.14) is not sufﬁcient to solve the problem. In addition to (1.14) a turbulence model is required that generates a relation between τ t+ (r) and du+ /dr to close the system of equations.

4

1 Fully Developed Turbulent Pipe Flow

1.2 Two-Layer Structure for Reτ → ∞ For high Reynolds numbers Reτ equation (1.14) reduces to τt+ = r

(core region) .

(1.15)

This equation is valid in the core region where the turbulent shear stress τt∗ is much larger than the viscous shear stress τv∗ . This equation, however, does not satisfy the condition τt+ = 0 at the wall which follows from the no-slip condition. Hence, (1.15) is not valid near the wall where τv∗ is not negligible any more. Hence, the ﬂow ﬁeld has a two-layer structure. It consists of the core region where (1.15) is valid and viscosity effects are negligible, and the wall layer where the viscous shear stress has the same order of magnitude as the turbulent shear stress. The two regions clearly have thicknesses of different order of magnitude. Whereas the thickness of the core region is of the order of magnitude of R ∗ , a wall-layer thickness δv∗ may be determined from the two characteristic quantities ν ∗ and u∗τ δv∗ =

R∗ ν∗ = . ∗ uτ Reτ

(1.16)

This tends to zero for high Reynolds numbers Reτ . The wall layer thickness is therefore small compared to R∗ for high values Reτ . Consequently, the ﬂow quantities in the wall layer are independent of R∗ in the leading order. In the following the velocity gradient will be found separately for the wall layer and for the core layer. The two solutions must be matched properly, i.e., they have to agree in an overlap layer.

1.3 Wall Layer In order to describe the wall layer it is suitable to introduce a dimensionless wall distance y ∗ u∗ y∗ (1.17) y + = ∗ = ∗ τ = Reτ (1 − r) . δv ν With this coordinate (1.14) changes into y+ du+ + + τ = 1 − . t dy + Reτ

(1.18)

Since the second term on the right side vanishes for high Reynolds numbers, the wall layer is a layer of constant total shear stress (equal to the wall shear stress), where the effect of the pressure gradient can be neglected. It is also independent of R ∗ , as mentioned earlier. Hence, the velocity u+ (y + ) is determined when τw∗ /ρ∗ and ν ∗ are given. It is therefore a universal function, called the universal law of the wall u+ (y + ). This universal law is valid not only for turbulent pipe ﬂow but also for all turbulent ﬂows with ﬁnite wall shear stress, cf. Schlichting and Gersten (2000).

1.4 Core Region

5

Some properties of this law are known a priori. The velocity gradient du + /dy + must become independent of the viscosity for large wall distance (in the overlap layer). This yields: 1 du+ = + κy y + →∞ dy + lim

(overlap layer) ,

(1.19)

where the constant κ is called Kármán constant after v. Kármán (1930). Integrating (1.19) leads to the well-known universal logarithmic law lim u+ (y + ) =

y + →∞

1 ln y + + C + κ

(overlap layer) .

(1.20)

As will be shown in Sect. 1.7, C + generally depends on the wall roughness. The no-slip condition and the continuity equation yield the asymptote close to the wall: du+ 3 = 1 − Ay + + . . . (1.21) lim + + y →0 dy lim u+ = y + −

y + →0

A +4 y + ... . 4

(1.22)

The following analytical description for the universal law of the wall satisﬁes the conditions (1.19) to (1.22); cf. Gersten and Herwig (1992), p. 378: 3

By + 1 du+ + = 3 dy + 1 + (A + B)y + 1 + κBy + 4

(1.23)

⎡ ⎤ + + 1 1 Λy π + 1 − 1 2Λy 1 ⎦ √ +√ + arctan u+ = ⎣ ln Λ 3 6 2 3 3 + + (Λy ) − Λy + 1 1 4 (1.24) + ln 1 + κBy + 4κ where

1 1 2π ln(κB) . + Λ = (A + B) 3 ; C + = √ 3 3Λ 4κ

(1.25)

The universal constants κ, C + and A can be determined from experiments, cf. Sect. 1.6.2. When κ, C + and A are known, the constants Λ and B follow from (1.25).

1.4 Core Region The distribution du+ /dr in the core region vanishes at the pipe axis (r = 0) and must agree with the equivalent wall layer distribution in the overlap layer. It is analogous to (1.19)

6

1 Fully Developed Turbulent Pipe Flow

1 du+ =− r→1 dr κ(1 − r)

(overlap layer) .

lim

(1.26)

If du+ /dr is known, this can be integrated to obtain the velocity distribution in the form of a velocity defect law as follows r u (r) − +

u+ c

=

du+ dr . dr

(1.27)

0 + In this equation u+ c is the velocity u on the axis or center (maximum velocity). To determine the velocity gradient, a turbulence modeling is necessary. Boussinesq (1872) proposed that, in analogy to Newton’s law of friction, see (1.10), the following ansatz should be used for τt∗ :

τt∗ = ρ∗ νt∗

du∗ dr∗

(1.28)

where νt∗ is called the (kinematic) eddy viscosity. Using dimensionless quantities according to (1.13) yields τt+ = −νt+

du+ dr

(1.29)

with the dimensionless eddy viscosity νt+ =

νt∗ . u∗t R∗

(1.30)

Combining (1.15) and (1.29) leads to the formula for the velocity gradient in the core region r du+ =− + . (1.31) dr νt Comparing (1.26) with (1.31) yields lim νt+ = κ(1 − r)

r→1

(overlap layer) .

(1.32)

The eddy-viscosity distribution νt+ (r) in the core region is a symmetrical function and must satisfy the condition (1.32). The following function has these properties and will be used as eddy-viscosity distribution: 1 − r2 1 + ar2 1 + br2 νt+ (r) = (1.33) κ 2(1 + a)(1 + b) or α β 2 κ = + + . (1.34) + 2 2 1−r 1 + ar 1 + br2 νt (r) These formulae have, beside the Kármán constant κ, the two free parameters a and b, which can be determined by comparison of the resulting velocity distribution with experimental data. The abbreviations α and β depend on a and b as follows:

1.4 Core Region

α=

2a2 (1 + b) a−b

,

β=

2b2 (1 + a) . b−a

7

(1.35)

Integration of (1.31) can be carried out easily because of (1.34) which leads to the velocity defect distribution β α 1 2 2 2 + + ln 1 + ar − ln 1 + br . (1.36) ln 1 − r − u (r) − uc = κ 2a 2b This distribution shows the logarithmic law in the overlap layer: lim u+ (r) = u+ c +

r→1

where C=

1 ln(1 − r) − C κ

(1.37)

β 1 α ln(1 + a) + ln(1 + b) − ln 2 . κ 2a 2b

(1.38)

In the overlap layer the two solutions (core region and wall layer) must be identical. The matching condition lim u+ (r) = lim u+ y + (1.39) r→1

y + →∞

combines (1.20) and (1.37) and leads because of (1.17) to a dependence of the maximum velocity on the Reynolds number Reτ : u+ c =

1 ln Reτ + C + + C . κ

(1.40)

The velocity averaged over the cross section of the pipe can be obtained by integration: 2 u∗m = ∗ ∗2 u+ m = u∗τ uτ R

∗

R

u∗ r∗ dr∗

0

1 u+ rdr = u+ c +C

=2

(1.41)

0

with 1 C = −2

+ uc − u+ (r) rdr

0

=−

(1.42)

β 1 α (1 + a) ln(1 + a) + 2 (1 + b) ln(1 + b) . κ 2a2 2b

The average velocity or bulk velocity shows the following dependence on Re τ when (1.40) and (1.41) are combined u+ m =

1 ln Reτ + C + + C + C . κ

(1.43)

8

1 Fully Developed Turbulent Pipe Flow

This is the friction law for turbulent pipe ﬂow at high Reynolds numbers Re τ in case that the pressure gradient (wall shear stress) is given and the volumen ﬂux (averaged velocity) is to be determined. It is (1.44) κ = 0.421 , C + + C + C = 2.55 according to the latest results of the Superpipe Experiments at Princeton University, see McKeon et al. (2003b). The velocity distribution according to (1.36) is shown in Fig. 1.2. 6

u*(r) - um* u* 2

0

-2

-4

-6

0

0.2

0.4

0.6

0.8

r

1

Fig. 1.2. Velocity defect distribution for fully developed pipe ﬂow according to (1.36), (1.40) and (1.43). (- - - -) Asymptote for r → 1 according to (1.37) and (1.41).

1.5 Friction Law for Given Volume Flux When the volume ﬂux is given the friction law is taken to be the dependence of the pipe friction factor λ=−

d∗ dp∗w /dx∗ 8 8τw∗ = − = 2 ρ∗ u∗m 2 ρ∗ u∗m 2 /2 u+ m

on the Reynolds number based on the pipe diameter d∗ 8 u∗m d∗ . = 2Reτ Re = ν∗ λ

(1.45)

(1.46)

Inserting λ and Re into (1.43) leads to the implicit form of the friction law for given Reynolds number Re

1.6 Velocity Distribution

√ 1 √ = C1 log Re λ − C2 λ

9

(1.47)

where ln 10 , C1 = √ 8κ

C2 = √

1 . ln 32 − 2κ C + + C + C 32κ

(1.48)

It is C1 = 1.934 ,

C2 = 0.554

(1.49)

by using the constants in (1.44). It is worth mentioning that these constants are valid in the range 3 × 105 ≤ Re ≤ 107 , see McKeon et al. (2003b). For Re < 3 × 105 , lower Reynolds number effects must be taken into account. This will be explained in Sect. 1.8. The data points in the Superpipe Experiments beyond Re = 10 7 have obviously been affected by roughness as will be discussed in Sect. 1.7. The explicit form of (1.47) is 2 κ G(Λ; D) (1.50) λ=8 ln Re with the G-function after Gersten and Herwig (1992), p. 782. This newly introduced function G(Λ; D) is deﬁned by Λ Λ + 2 ln −D =Λ (1.51) G G and is tabulated in the mentioned reference. It satisﬁes the asymptotic condition lim G(Λ; D) = 1 .

Λ→∞

(1.52)

In the application here, it was set Λ = 2 ln Re ,

. D = 2 ln κ + κ C + + C + C

(1.53)

It is D = 0.417 by using (1.44).

1.6 Velocity Distribution 1.6.1 General Formula After the velocity distributions have been determined for the two regions in the ﬂow + + ﬁeld, i.e., the u+ WL (y ) according to (1.24) for the wall layer and u CR (r) according to (1.36) for the core region, a composite solution can be constructed, cf. Van Dyke (1975). The so-called additive composition is the sum of these two solutions which is corrected by subtracting the part they have in common, i.e., the solution in the overlap layer according to (1.20):

10

1 Fully Developed Turbulent Pipe Flow

+ 1 + − ln y + − C + u+ (r, Reτ ) = u+ CR (r) + uWL y κ

(1.54)

where y + and r are connected by (1.17). When (1.24), (1.36) and (1.40) are used, the composite velocity distribution has the form: β 1 α 2 2 ln 1 + ar − ln 1 + br +C u (r) = ln(1 + r) − κ 2a 2b 1 Λy + + 1 2Λy + − 1 π 1 1 √ + +√ ln arctan + Λ 3 6 2 3 3 (Λy + ) − Λy + + 1 +

+

1 4 ln 1 + κBy + 4κ (1.55)

√ Re λ (1.56) y = √ (1 − r) . 2 8 The average velocity according to (1.41) has been determined in (1.42) by taking into account only the velocity distribution in the core region. When the integration is carried out over the composite velocity distribution, the relevant constant C turns out to be C = C CR + ∆C . (1.57)

where

+

Here C CR is the constant in (1.42), when only the velocity distribution in the core region is taken into account. The effect of the velocity distribution in the wall layer is given by the correction term ∆C, for which it follows 2 ∆C = − Reτ

∞ 0

+ 1 + + + ln y + C − uWL y dy + . κ

(1.58)

(y ) with the constants κ = 0.421; C + = 5.60; A = 6 × 10−4 Using (the values of C and A are explained later) leads to (1.24) for u+ WL +

+

∆C = −

102 , Reτ

(1.59)

see also Gersten and Herwig (1992), p. 630. For Reynolds numbers higher than Re = 106 (Reτ > 2378) the correction ∆C results in a reduction of u+ m according to (1.43) which is smaller than 0.2 percent. Only for Reynolds numbers less than Re = 3 × 105 the correction ∆C should be taken into account, see also Sect. 1.8.

1.6 Velocity Distribution

11

1.6.2 Determination of the Model Constants The velocity distribution (1.55) contains ﬁve free constants κ, C + , a, b, and A. These can be determined by comparison with experiments as follows: 1. From the friction law, (1.43) or (1.47), κ and C + + C + C can be found. 2. The velocity defect distribution (1.36) shows a quadratic behavior near the axis (1 + a)(1 + b) 2 2+α+β 2 r =− r . lim u+ (r) − u+ c =− η→0 2κ κ

(1.60)

Note 1.2 (Alternative for (1.60)) Instead of using (1.60) one could take the location

where u+ = u+ m as one condition to determine the model constants. Experiments at high Reynolds numbers show for this location r ≈ 0.75, see Fig. 1.2.

3. The combination of (1.35), (1.41) and (1.42) results in + u+ c − um = −C =

(1 + a)(1 + b) 1 + a ln . κ(a − b) 1+b

(1.61)

When κ has been determined via the friction law, the constants a and b can be calculated from (1.60) and (1.61). Then C follows from (1.38) and C + from the combination C + + C + C, found via the friction law. 4. The ﬁfth constant A is given by A=−

d4 u+ 1 lim 6 y+ →0 dy + 4

(1.62)

because of (1.22). When (1.24) is applied for the law of the wall, A is connected with the velocity u+ at y + = 15 by A=

u+ (y + = 15) − 4.4 −4 10 . 1.07

(1.63)

The newest results of the Superpipe Experiments at Princeton University yield κ = 0.421 , C + + C + C = 2.55 , C = −4.28 , C

+

= 5.60

(1.64)

,

where C + was determined by using (1.20). From these constants it follows C = 1.23 ,

(1.65)

which is in very good agreement with the Superpipe Experiments by using (1.37), although McKeon et al. (2004) found the slightly different value C = 1.20 ± 0.1.

12

1 Fully Developed Turbulent Pipe Flow

By applying the values κ, C and C, (1.38) and (1.42) lead to a = −0.2714

b = 5.567

(1.66)

β = 7.735 .

(1.67)

and via (1.35) to α = −0.1656

These constants have been applied to calculate the universal velocity distribution for high Reynolds numbers according to the combination of (1.36) and (1.41). The result is shown in Fig. 1.2. The constant A can be found in the literature as a number in the interval 5 ≤ A × 104 ≤ 7, see Gersten and Herwig (1992), p. 377. Numerous measurements have brought out u+ (y + = 15) = 10.6 cf. Reichardt (1951) and Kestin and Richardson (1963). At higher Reynolds number, this value is only slightly higher. McKeon et al. (2003a) found u+ (y + = 15) = 10.7. Therefore applying (1.63) leads to the rough estimation (1.68) A = 6 × 10−4 . Note 1.3 (The Tragicomedy of Fluid Mechanics) Fluid mechanics is supposed to be a great science. Outstanding progress in theoretical, numerical and experimental ﬂuid mechanics has been made in the last decades. Ironically enough, however, for the simplest technical ﬂow one can think of, namely the incompressible, fully developed turbulent ﬂow in a straight circular pipe, a reliable friction law is not available yet. Millions of such ﬂows have been realized under extremely precise conditions in the numerious national Ofﬁces of Weights and Measures or Bureau of Standards but, unfortunately, only the volume ﬂow has been measured, not simultaneously the pressure gradient. The opposite case, measuring the pressure gradient, but not the volume ﬂow, is also available, see AGARD-AR-345 (1998), Case PCH02. As a result only a few experimental data for the friction law are available. The uncertainty, however, is unacceptable. In the recent collection of test data in the AGARD-Report just mentioned (selected for the validation of Large-Eddy-Simulations!) the most important Kármán constant κ varies in a range between κ = 0.39 and κ = 0.436. It is hoped that the constants, chosen in this chapter, will bring us very close to the truth. Note 1.4 (Power-Law Velocity Distributions) Sometimes the velocity distributions of fully developed turbulent pipe ﬂows have been modeled by power laws. For example, Nikuradse (1932) tried power law representations and found a dependence of the exponents on the Reynolds number. But he abandoned this mode of correlation as inferior to the log-law representation described in this chapter. Recently, Barenblatt (1993) reviewed the power-law model and rationalized the form of the power law α(Re) (1.69) u+ = B(Re) y + by using scaling arguments, see also Barenblatt et al. (1997a, b). On the basis of the data of Nikuradse (1932) he proposed expressions for B(Re) and α(Re) as functions of the Reynolds number. Unfortunately, the power-law model for the velocity distribution is seriously in error near the pipe axis and close to the pipe wall. These errors might have been anticipated from the prediction by (1.69) of an unbounded velocity gradient at the wall and a ﬁnite velocity gradient at the centerline. Afzal (2001) has shown that power law and log-law velocity proﬁles

1.7 Inﬂuence of Wall Roughness

13

are equivalent when the power law velocity distribution is restricted to the overlap layer, see also Barenblatt and Chorin (1998). The drawback of the power-law model is again that the matching condition in the overlap layer leads to a function of the Reynolds number rather than to the universal Kármán constant. For practical applications, the log-law model is obviously much more attractive than the power-law model, see also Churchill (2001).

1.7 Inﬂuence of Wall Roughness 1.7.1 Hydraulically Deﬁned Roughness So far in the analysis of this chapter it has been assumed that the inner pipe wall is smooth. It turns out that everything is valid for rough walls as well, with one exception: the integration constant C + according to (1.20) now depends on the wall roughness. From dimensional analysis it follows that C + (k + ) is a function of the dimensionless roughness parameter (sometimes called roughness Reynolds number) k+ =

k ∗ u∗τ . ν∗

(1.70)

Here k ∗ is the roughness height or simply roughness. It can be interpreted geometrically in connection with the well-known experiments by Nikuradse (1933). For his investigations on artiﬁcially roughened pipes, Nikuradse coated the inside of the pipes with carefully graded sand grains, glued in place. He used the mesh size of the grading screens as roughness k ∗ . Therefore, k ∗ is also called sandgrain roughness or Nikuradse roughness. The roughness Reynolds number k + is the ratio of the Nikuradse roughness k ∗ and the wall-layer thickness δv∗ deﬁned in (1.16). It is plausible that the viscosity effect must disappear completely when k ∗ δv∗ or k + → ∞. In this case, the friction factor must become independent of the viscosity. This so-called fully rough pipe ﬂow is governed by Kármán’s law 1 3.71d∗ 8 + = ln . (1.71) um = λ κ k∗ Inversion of this law leads to √ √ k∗ = 3.71 exp − 8κ/ λ , d∗

(1.72)

which deﬁnes the roughness k ∗ . For a given condition of the internal wall, each pipe has a roughness k ∗ . However, k ∗ is not deﬁned in a geometrical but rather in a hydraulical sense. It can be determined via (1.72) by measuring the friction factor λ at a high enough Reynolds number such that fully rough pipe ﬂow is guaranteed. In principle, λ must be measured at several different Reynolds numbers to ensure that it becomes independent of the Reynolds number. Combining (1.43) and (1.71) leads to

14

1 Fully Developed Turbulent Pipe Flow

+ 2.004 1 + − C − C − ln k + . lim C k = κ κ k+ →∞

(1.73)

When the constants of the Superpipe Experiments (1.64) are used, the limits of the function C + (k + ) are: lim C + (k + ) = 5.60

k+ →0

lim C + (k + ) = 7.81 −

k+ →∞

1 ln k + κ

hydraulically smooth

(1.74)

fully rough

(1.75)

It should be mentioned that k ∗ according to (1.72) depends on the Kármán constant κ being used. Hence, any roughness k ∗ determined with κ = 0.407 is not directly comparable with the Nikuradse roughness (κ = 0.407) as it is usually found in the literature. This, however, is not so detrimental because the friction factor is not very sensitive with respect to the roughness. Now the following question arises: What is the correct formula for the function C + (k + ) in the transition regime between smooth ﬂow and fully rough ﬂow? Such a formula must satisfy the two limiting cases for smooth ﬂow (k + → 0) and fully rough ﬂow (k + → ∞). Roughness may be classiﬁed in various ways. For instance, roughness elements may be discrete or distributed and their geometry may be deterministic (regular) or stochastic (random). Experiments have shown that for many roughness geometries a single parameter is inadequate to specify the roughness effect on the ﬂow. Parameters such as the roughness height, shape, density and manner of distribution may be important in determining the net effect of the roughness on the pipe ﬂow, see Waigh and Kind (1998). For all possible roughness geometries, the so-called equivalent sandgrain roughness can be determined by using (1.72) in fully rough ﬂow. But various roughness geometries show in general a different behavior in the transition regime between smooth and fully rough ﬂow. 1.7.2 Natural Roughness Colebrook (1938/1939), in collaboration with White [see (Colebrook and White, 1937)] found that the parameter k + is sufﬁcient to describe most forms of naturally rough commercial pipes in the transition regime. Therefore, in the following only naturally rough commercial pipes with the tech + nical roughnesss will be considered. According to Colebrook the function C + ktech is a simple combination of the two limiting cases mentioned above 1 + ln 2.54 + ktech κ k+ 1 = 5.60 − ln 1 + tech . κ 2.54

+ ) = 7.81 − C + (ktech

(1.76)

1.7 Inﬂuence of Wall Roughness

15

+ in (1.47) and using the constants (1.64) leads Inserting this formula for C + ktech to the following friction law for rough commercial pipes ∗ ∗ k /d 1.934 1 √ √ = −1.934 log + . (1.77) 3.71 λ Re λ In the limiting case k ∗ /d∗ → 0 or k + → 0 at ﬁnite Reynolds number, (1.77) reduces to (1.47). In the other limiting case Re → ∞ at ﬁnite k ∗ /d∗ or k + → ∞ (1.77) is reduced to (1.71). + The roughness function (1.76) after Colebrook-White leads for small values k tech to + + lim C + (ktech ) = 5.60 − ktech /(2.54κ) . (1.78) + ktech →0

However, Bradshaw (2000) has shown that the correct form is + +2 = 5.60 − α ktech . lim C + ktech + ktech →0

(1.79)

The coefﬁcient α for commercial pipes is not known yet. But it can be estimated from the Superpipe Experiment, cf. Zagarola and Smits (1998) and McKeon et al. ∗ = 0.45 µm (2003b). This particular pipe (d∗ = 0.129 m) had the roughness ktech + 7 and the ﬂow was hydraulically smooth for Re ≤ 10 or for ktech ≤ 1. The correct smoothness condition can be derived from (1.43) and (1.79) √ 14 ∆λ 2 1 ∆λ +2 √ ≈√ (1.80) ktech ≤ α λ λ λ λ when α ≈ 0.1√was chosen so that (1.80) is valid for the Superpipe Experiments (Re = 107 , 1/ λ = 11, ∆λ/λ = 0.00005/0.0082 = 0.006, k + ≤ 1). This formula shows clearly that the smoothness criterion for surfaces with sandgrain roughness, k + ≤ 5, cannot be applied for natural roughness of commercial pipes. The behavior of these two roughness types have a quite different behavior in the transition regime between hydraulically smooth and fully rough ﬂows. The roughness effects can therefore be divided into the following three regimes ⎫ √ + ⎪ 14/ λ ∆λ/λ hydraulically smooth 0 ≤ ktech ≤ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ √ √ + 14/ λ ∆λ/λ < ktech < 420 λ/ ∆λ/λ ⎪ (1.81) transition region ⎪ ⎪ ⎪ ⎪ ⎪ √ ⎭ + fully rough 420 λ/ ∆λ/λ ≤ ktech Here ∆λ/λ is the relative uncertainty of the friction factor. 1.7.3 Velocity Distribution The velocity defect distribution for rough walls is the same as for smooth walls, see (1.36). But the maximum velocity u+ according to (1.40) is reduced by roughness

16

1 Fully Developed Turbulent Pipe Flow

because of (1.76). Therefore, the rough-wall velocity proﬁle when plotted in walllayer coordinates (u+ , y + ) is shifted downward by a constant amount + + + + ∆u+ = u+ smooth − urough = C (0) − C (k ) + ktech 1 8 8 − = ln 1 + = λsmooth λrough κ 2.54

(1.82)

Reτ =const

+ is called roughness function. The function ∆u+ ktech If the surface is rough, it is difﬁcult to determine the origin of the coordinate system y = 0. It is common practice to choose this origin so that the overlap law (1.20) is satisﬁed, cf. Grigson (1984), Acharya et al. (1986). 1.7.4 Geometric Roughness Parameters To describe the geometric proﬁle of a rough surface, the following geometric roughness parameters are in use: a) Arithmetic mean roughness Ra∗

1 = ∗

∗

∗ |y ∗ (x) − ym | dx∗

(1.83)

0

∗ where y ∗ (x∗ ) describes the roughness proﬁle and ym is the mean of y ∗ (x∗ ) :

1 ∗ ym = ∗

∗

y ∗ (x∗ ) dx∗ .

(1.84)

0

b) Root-mean-square roughness ∗ 1 ∗ ]2 dx∗ Rq∗ = ∗ [y ∗ (x∗ ) − ym

(1.85)

0

∗ This value is also written as Rrms .

c) Maximum proﬁle valley depth ∗ ∗ ∗ Rt∗ = Rmax = ymax − ymin

for a given reference length ∗ .

(1.86)

1.7 Inﬂuence of Wall Roughness

17

d) Average maximum proﬁle valley depth (also: ten point height of irregularities) 1 ∗ ∗ − ymin )i (y 5 i=1 max 5

Rz∗ =

(1.87)

In this case the reference length ∗ is divided into ﬁve lengths ∗i = ∗ /5 and ∗ ∗ for each length ∗i the difference between the peak ymax and the valley ymin is determined. Table 1.1. Examples of the ratio of the geometric roughness parameters Ra∗ , Rq∗ , Rz∗ , Rt∗ for surfaces of commercial steel pipes. Effects of painting, sanding, polishing and coating. Ra∗ / µm

Rq∗ /Ra∗

Rz∗ /Ra∗

Rt∗ /Ra∗

8.5

1.3

6.8

8.0

Schultz (2002)

0.4 – 2.7

1.8

14

16

steel painted sanded polished

Schultz (2002)

0.18

1.7

11

12

steel coated

Gersten et al. (2000)

3.9

1.3

6.3

7.9

Surface

Reference

steel

Gersten et al. (2000)

steel painted unsanded or sanded

Table 1.1 shows examples of geometric roughness parameters and their ratios. The geometric roughness parameters can be determined by using mechanical proﬁlometers. Laser interferometry can also be applied which leads to about 12% higher roughness parameters. 1.7.5 Relation Between Hydraulic and Geometric Roughness Parameters There is general agreement that a relation between the hydraulic roughness parameter ∗ ktech and the geometric roughness parameters in (1.83) to (1.87) exists. But only very few investigations have led to quantitative results for this relation. The results of investigations of the natural roughness of commercial pipes are summarized in Table 1.2. From these data it follows that the natural roughness of commercial pipes is approximately ∗ = 3.5Ra∗ . (1.88) ktech It is worth mentioning that in DIN EN ISO 5167-1 (January 2004) the relation ∗ ∗ ktech = πRa∗ is used. The correlation of ktech with other geometric roughness parameters can be estimated by the results in Table 1.1.

18

1 Fully Developed Turbulent Pipe Flow

∗ of commercial pipes and geometric roughTable 1.2. Relationship between roughness ktech ness Ra∗ according to various authors.

Reference

∗ ktech /Ra∗

Speidel (1962) Uhl (1965) Acharya et al. (1986) Revell (1998) Zagarola and Smits (1998)

3.1 3.3 4.2 3.1 3.9

1.8 Low Reynolds Number Pipe Flow In the previous Sections pipe ﬂows at high Reynolds numbers have been treated. The empirical constants according to the Superpipe Experiment in (1.44) and (1.64) turned out to be valid only for about Re ≥ 3 × 105 . In this Section, turbulent pipe ﬂows for Re < 3 × 105 will therefore be dicussed. Pipe ﬂows for Reynolds numbers below about Re = 2000 are laminar. For pipe Reynolds numbers between 2000 and about 4000 the friction factor can have large uncertainties and is essentially indeterminate. This region is, therefore, called critical zone, cf. Benedict (1980). Detailed experimental investigations, for instance by Rotta (1956), have shown that the ﬂow in this zone has an intermittent character. This means that the ﬂow is occasionally laminar and occasionally turbulent. Hence, the velocity distribution alternates between a corresponding fully developed laminar and a corresponding fully developed turbulent ﬂow. However, for Re > 4000 the pipe ﬂow becomes steady and reasonably determinate. It will be therefore discussed next, how the analysis presented so far can be extended to lower Reynolds numbers in the range 4000 < Re < 3 × 105 . As already mentioned the analysis in the previous Sections referred to pipe ﬂows at very high Reynolds numbers. These solutions can be extended systematically to lower Reynolds numbers. For that purpose the solutions are set as asymptotic series where = 1/Reτ is the perturbation parameter. The leading terms of these series are the solutions presented so far. The terms that follow in the asymptotic series represent the higher order terms which describe the deviations from the known asymptotic solutions at lower Reynolds numbers. Details of the procedure for such singular perturbation problem are given in Gersten and Herwig (1992). The extension of the friction-factor formula (1.43) to lower Reynolds number is as follows: 1 1 1 + ln Reτ + C2 + . . . (1.89) um = ln Reτ + C1 + κ Reτ κ2 where

1 = C + + C + C . C

(1.90)

2 . The correction term The higher order term contains two more constants 1/κ2 and C (1.59) is included in C2 . The new constants can again be determined by comparison

1.8 Low Reynolds Number Pipe Flow

19

with experimental data. Afzal and Yajnik (1973) have made such an adjustment and found: 1/κ = 2.5

1 = 1.28 C

;

;

1/κ2 = 0

;

2 = 75 C

(1.91)

see also Afzal (1976). Since nowadays experimental data on pipe ﬂows over a wider range of Reynolds numbers are available, the numbers in (1.91) must obviously be revised. When (1.89) is written in the form 1 1 1 1 + C2 ln Reτ + C1 + (1.92) + um = κ Reτ κ2 Reτ it is clear that the function u+ m (Reτ ) plotted in a semi-logarithmic diagram is in general not any more a straight line. In order to determine the two new constants in the higher-order term, reliable data in the lower Reynolds number range must be applied. In AGARD-AR-345 (1998) a selection of test cases of turbulent pipe ﬂows have been collected for the validation of computer codes. For the following two test cases sufﬁcient information is available: 1. Durst et al. (1995): Re = 7442

,

Reτ = 250

,

√ 1/ λ = 5.26

(1.93)

,

Reτ = 691

,

√ 1/ λ = 6.29

(1.94)

2. den Toonder (1995): Re = 24580

It is worth mentioning that this data point satisﬁes exactly the well-known friction law by Blasius (1913). Taking these data and those of the Superpipe Experiments, cf. McKeon et al. (2004), the following constants can be determined: 1 = 2.55 C

2 = −201 . (1.95) C √ This leads to the friction law for the so-called transmission factor 1/ λ valid for all turbulent pipe ﬂows above Re = 4000 (hydraulically smooth): 1/κ = 2.375

or

;

;

1/κ2 = 0

;

71 1 √ = 0.8398 ln Reτ + 0.9016 − Reτ λ

(1.96)

√ 1 402 √ √ = 1.934 log Re λ − 0.554 − λ Re λ

(1.97)

The comparison of the formula (1.96) with the experimental data is shown in Fig. 1.3. It should be mentioned that Zagarola and Smits (1998) found empirically √ 0.9 , which can be considered as the higher-order term in (1.97) as −233/ Re λ a conﬁrmation of the analysis presented here. It is worth noting that Wosnik et al. (2000) worked out a slightly different analysis, but ended up with a similar higherorder term proportional to (ln Reτ )−0.441 .

20

1 Fully Developed Turbulent Pipe Flow 13

4 3 4.10 10

10

5

10

6

10

7

Re

1 11 10 9 8 7 6 5 4 2 10

10

3

10

4

10

5

Re

106

Fig. 1.3. Friction law for turbulent pipe ﬂows for Re> 4000 (smooth surface). 8/λ; (—–) Formula (1.96); (- - - -) Asymptote for Re → ∞ according to (1.43), u+ m = () Experiment by Durst et al. (1995); (2) Experiment by den Toonder (1995); (◦) Experiment by McKeon et al. (2004).

References Acharya M, Bornstein J, Escudier MP (1986) Turbulent boundary layers on rough surfaces. Experiments in Fluids 4:33–47 Afzal N (1976) Millikan’s argument at moderately large Reynolds number. Phys. Fluids 19:600–602 Afzal N (2001): Power law and log law velocity proﬁles in fully developed turbulent pipe ﬂow: equivalent relations at large Reynolds numbers. Acta Mechanica, Vol. 151:171–183 Afzal N, Yajnik K (1973) Analysis of turbulent pipe and channel ﬂows at moderately large Reynolds number. J. Fluid Mech. 61:23–31 AGARD (1998) A selection of test cases for the validation of large-eddy simulations of turbulent ﬂows. AGARD Advisory Report 345 Barenblatt GI (1993) Scaling laws for fully developed turbulent shear ﬂows. Part I: Basic hypothesis and analysis. J. Fluid Mech. 248:513–520 Barenblatt GI, Chorin AJ (1998) Scaling of the intermediate region in wall-bounded turbulence: The power law. Phys. Fluids, Vol. 10:1043–1044 Barenblatt GI, Chorin AJ, Prostokishin VM (1997a) Scaling laws for fully developed turbulent ﬂow in pipes. Appl. Mech. Rev., Vol. 50:413–429 Barenblatt GI, Chorin AJ, Prostokishin VM (1997b): Scaling laws for fully developed turbulent ﬂow in pipes: Discussion of experimental data. Proc. Natl. Acad. Sci. USA, Vol. 94, 773 Benedict RP (1980) Fundamentals of Pipe Flow. John Wiley Sons, New York Blasius H (1913) Das Ähnlichkeitsgesetz bei Reibungsvorgängen in Flüssigkeiten. Forschg. Arb. Ing.-Wesen, Heft 131, Berlin Boussinesq J (1872) Essai sur la theorie des eaux courantes. Memoires Acad. des Sciences, Vol. 13, No. 1, Paris

References

21

Bradshaw P (2000) A note on “critical roughness height” and “transitional roughness”. Phys. Fluids 12:1611–1614 Churchill SW (2001) Turbulent ﬂow and convection: the prediction of turbulent ﬂow and convection in a round tube. In: Advances in Heat Transfer, Academic Press, San Diego, Vol. 34:255–361 Colebrook CF (1938/1939) Turbulent ﬂow in pipes with particular references to the transition region between the smooth and the rough pipe laws. J. Inst. Civil Eng., London, 11:133–156 and 12:393–422 Colebrook CF, White CM (1937) Experiments with ﬂuid friction in roughened pipes. Proc. Royal Soc., London, Series A 161:367–381 den Toonder JMJ (1995) Drag reduction by polymer additives in a turbulent pipe ﬂow: Laboratory and numerical results. Ph. D. thesis, Delft University of Technology Durst F, Jovanovic J, Sender J (1995) LDA measurements in the near-wall region of a turbulent pipe ﬂow. J. Fluid Mech. 295:305–335 Gersten K, Herwig H (1992) Strömungsmechanik - Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig, Wiesbaden Gersten K, Papenfuss HD, Kurschat T, Genillon P, Fernández Pérez F, Revell N (2000) New transmission-factor formula proposed for gas pipelines. Oil & Gas Journal, February 14 Grigson CWB (1984) Nikuradse’s experiment. AIAA Journal 22:999–1001 Kestin J, Richardson PD (1963) Heat transfer across turbulent incompressible boundary layers. Int. J. Heat Mass Transfer 6:147–189. Also: Forsch. Ing.-Wesen 29:93–104 McKeon BJ, Li J, Jiang W, Morrison JF, Smits A (2003a) Pitot probe corrections in fullydeveloped turbulent pipe ﬂow. Meas. Sci. Technol. 14:1449–1458 McKeon BJ, Morrison JF, Jiang W, Li J, Smits AJ (2003b) Revised log-law constants for fully-developed turbulent pipe ﬂow. In: AJ Smits (Ed.): IUTAM Symposium on Reynolds Number Scaling in Turbulent Flow. Kluwer Academic Publishers, Dordrecht McKeon BJ, Li J, Jiang W, Morrison JF, Smits AJ (2004): Further observations on the mean velocity in fully-developed pipe ﬂow. J. Fluid Mech. 501:135–147 Nikuradse J (1932) Gesetzmäßigkeiten der turbulenten Strömung in glatten Rohren. VDIForsch. Arb. Ing.-Wesen Nikuradse J (1933) Strömungsgesetze in rauhen Rohren. VDI-Forsch.-Heft 361; VDI-Verlag, Berlin Revell N (1998) Internal Report of BG Technology, PR 091 Reichardt H (1951) Vollständige Darstellung der turbulenten Geschwindigkeitsverteilung in glatten Leitungen. Z. angew. Math. Mech. 31:203–219 Rotta JC (1956) Experimenteller Beitrag zur Entstehung turbulenter Strömung im Rohr. Ing. Arch. 24:258–281 Schlichting H, Gersten K (2000) Boundary-Layer Theory. Springer-Verlag, Berlin, Heidelberg; 8th Edition, Corrected Printing 2003 Schultz MP (2002) The relationship between frictional resistance and roughness for surfaces smoothed by sanding. J. Fluids Eng. 124:492–499 Speidel L (1962) Determination of the necessary surface quality and possible losses due to roughness in steam turbines. Elektrizitätswirtschaft 61:799–804 Uhl AE (1965) Steady ﬂow in gas pipelines. American Gas Association. Technical Report No. 10 Van Dyke M (1975) Perturbation Methods in Fluid Mechanics. The Parabolic Press, Stanford, California

22

1 Fully Developed Turbulent Pipe Flow

von Kármán Th (1930) Mechanische Ähnlichkeit und Turbulenz. Nachr. Ges. Wiss. Göttingen. Math. Phys. Klasse: 58-76 und Verhandlg. des III. Intern. Kongresses für Techn. Mechanik, Stockholm, Teil I:85–93 Waigh DR, Kind RJ (1998) Improved aerodynamic characterization of regular threedimensional roughness. AIAA J. 36: 1117-1119 Wosnik M, Castillo L, George WK (2000) A theory for turbulent pipe and channel ﬂows. J. Fluid Mech. 421:115-145 Zagarola MW, Smits AJ (1998) Mean-ﬂow scaling of turbulent pipe ﬂow. J. Fluid Mech. 373:33–79

2 Decay of Disturbances in Turbulent Pipe Flow Klaus Gersten, Heinz-Dieter Papenfuss Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

The basic equations for turbulent ﬂows in circular pipes at high Reynolds numbers are derived. At a certain cross section a three-dimensional velocity ﬁeld is prescribed that differs only slightly from the fully developed state. The axial decay of the disturbances is investigated by solving the basic equations using an asymptotic expansion of the solutions. The resulting system of linear ordinary differential equations leads to eigenvalue problems. Eigenvalues and eigenfunctions are presented. The general solution can be expanded into series of eigensolutions. The decaying ﬂow consists of various ﬂow structures that fall into two categories: the U -type (ring vortex, sourcesink pair, source-sink quadrupole, etc.) and the W -type (single, double and multiple streamwise vortices). Each of these ﬂow structures has its own decay behavior. Swirl (single streamwise vortex) has the lowest decay followed by the source-sink pair. The strength of the individual ﬂow structure is described by means of characteristic parameters that are the coefﬁcients in the series expansions of the solution. Comparisons of the theory with experimental results from the literature show very good agreement. The different dependence of the decay rate on the Reynolds number for ﬂows with and without swirl can be clariﬁed. The theory is valid for smooth as well as for rough pipe walls.

2.1 Introduction Flows in straight pipes are often not fully developed. Deviations from the fully developed state may occur due to bends, valves or other pipe ﬁttings. Even swirl can be found downstream of a pair of bends installed in perpendicular planes. The decay of the deviations from the fully developed condition is of great practical importance. For example, these disturbances can cause signiﬁcant errors in the measurement of the mass ﬂow rate. It is therefore important to know which length of a straight pipe is required for a given upstream ﬂow disturbance to reduce to a speciﬁed acceptable deviation from the fully developed ﬂow condition. It is also worth knowing the additional pressure drop due to the disturbances.

24

2 Decay of Disturbances in Turbulent Pipe Flow

Experimental as well as theoretical investigations on such disturbed turbulent pipe ﬂows have been carried out to a great extent. The simplest case is the axisymmetric ﬂow without swirl as it occurs, for instance, in the ﬁnal stage of a turbulent pipe inlet ﬂow. The experimental studies by Klein (1981), Barbin and Jones (1963) and Laws et al. (1987) and the theoretical work by Cebeci and Chang (1978) and Voigt (1995) are worth mentioning. More recent investigations on axisymmetric pipe ﬂow with swirl have been carried out by Kitoh (1991), Parchen (1993), Reader-Harris (1994), Steenbergen (1995) and Steenbergen and Voskamp (1998). The general case of three-dimensional disturbances in the pipe has been studied only in a few experimental investigations, for instance by Norman et al. (1989), Sudo et al. (1998), Kalkühler (1998) and Mickan (1999). Theoretical studies with general results on the decay of three-dimensional disturbances in a pipe are not available apart from the work by Gersten and Klika (1998) for the three-dimensional laminar case. In order to quantify the intensity of various ﬂow disturbances in the pipe certain characteristic parameters (sometimes denoted as “ﬂow indices”) are used. The axisymmetric ﬂow without swirl can be characterized by the blockage factor, which is essentially the ratio between the maximum velocity and the average velocity; see Klein (1981). The swirl intensity can be determined by the swirl number, which is a dimensionless radial moment of the circumferencial momentum ﬂux. Corresponding characteristic parameters can be deﬁned for three-dimensional disturbances; see Sudo et al. (1998) and Mickan (1999). Many experiments have shown that these characteristic parameters decay downstream in form of an exponential function: S(x) = S0 exp(−αx/D), where S stands for the swirl number or any other characteristic parameter describing the ﬂow disturbances. Experimental as well as theoretical investigations showed a dependence of α on the Reynolds number. With respect to swirl, α turned out to be proportional to the friction factor λ of the fully developed pipe ﬂow according to Norman et al. (1989), Reader-Harris (1994), Steenbergen (1995) and Steenbergen and Voskamp (1998). For the axisymmetric disturbance without swirl, however, the coefﬁcient α was found to be proportional to the square-root of the friction factor λ according to Herwig and Voigt (1995) and Voigt (1995). The same behavior was found for the three-dimensional disturbance downstream of an elbow (essentially two counterrotating longitudinal vortices) according to Norman et al. (1989). The purpose of the present investigation is to treat general three-dimensional disturbances in turbulent pipe ﬂow in a similar way as it was done by Gersten and Klika (1998) for the laminar case. Of particular interest is the correct description of the decaying of the various disturbances as function of the Reynolds number.

2.2 Basic Equations 2.2.1 Two-Layer Structure The turbulent incompressible steady ﬂow in a straight circular pipe is considered. At high Reynolds numbers the turbulent pipe ﬂow consists of two regions: 1) the core

2.2 Basic Equations

25

region where the viscous effects are negligible compared to the turbulent momentum transfer and 2) the viscous wall layer. The latter is characterized by a universal untwisted velocity distribution depending on the local shear stress τw∗ and the viscosity ν ∗ ; see Schlichting and Gersten (2000). This is in complete analogy to the two-layer structure of turbulent three-dimensional boundary layers as described in the textbook by Schlichting and Gersten (2000) on page 638. The basic equations for the core region are therefore free of viscous effects. The boundary conditions for these equations are not the conditions at the pipe wall, but rather the matching conditions with the wall-layer solution, i.e., the well-known logarithmic velocity distribution in the overlap layer of the core region and wall region. 2.2.2 Core Region The basic equations in the core region in vector form are as follows; see Schlichting and Gersten (2000), page 73. Continuity equation (2.1) div w ∗ = 0 momentum equation 1 1 ∗2 w ∗ = − ∗ grad p ∗ + Div τt∗ −w ∗ × ω grad 2 ρ where

∗ ω ∗ = curl w

(2.2)

(2.3)

is the vorticity and τt∗ the turbulent stress tensor. Dimensional quantities are marked by an asterisk. A cylindrical coordinate system is used, see Fig. 2.1. The following

R* r,v ,w

x,u

Fig. 2.1. Geometry and coordinate system.

26

2 Decay of Disturbances in Turbulent Pipe Flow

dimensionless coordinates are introduced: x=

x∗ L∗

r=

r∗ R∗

(2.4)

where R∗ is the pipe radius and L∗ is a length in axial direction characteristic for the decaying of the ﬂow disturbances considered. The ratio 1 =

R∗ L∗

(2.5)

is assumed to be a small number. The axial velocity component is based on the mean velocity or bulk velocity 1 u∗m = π R∗ 2

2π R∗ u∗ (x∗ , r∗ , θ)r∗ dr∗ dθ . 0

(2.6)

0

To avoid a degeneration of the continuity equation, the velocity components v ∗ and w∗ are based on 1 u∗m . In other words, these velocity components are small compared with u∗m . An eddy-viscosity concept is applied to the turbulent stress tensor. Therefore, the basic equations for the fully turbulent core region are the same as those for laminar ﬂow, when the viscosity µ∗ is replaced by the eddy viscosity µ∗t . Hence, the following dimensionless quantities are used: u=

u∗ u∗m

v=

v∗ 1 u∗m

νt+ =

w=

w∗ 1 u∗m

p=

1 µ∗ µ∗t . = ∗ ∗t ∗ ∗ ∗ ∗ ρ R uτ ρ R uτ ∞ γ

p∗ − p∗0 , ρ∗ u∗2 m

(2.7)

The eddy viscosity is based on the local skin-friction velocity u ∗τ deﬁned by the (negative) local resultant wall shear stress τw∗ (x∗ , θ) = −ρ∗ u∗τ 2 (x∗ , θ) .

(2.8)

The skin-friction velocity u∗τ (x∗ , θ) tends to the value u∗τ ∞ = u∗τ (x∗ → ∞; θ) of the fully developed ﬂow far downstream. The function γ (x∗ , θ) =

u∗τ (x∗ , θ) u∗τ ∞

(2.9)

tends to 1 for x∗ → ∞. From the asymptotic theory for the fully developed turbulent pipe ﬂow it is wellknown that the value 1 λ u∗τ ∞ =O (2.10) 2 = ∗ = um 8 ln Re

2.2 Basic Equations

27

is a small number, too, when the Reynolds number Re =

ρ∗ u∗m 2R∗ µ∗

(2.11)

becomes large. The friction factor λ is a measure of the friction resistance in the fully developed pipe ﬂow. In principle, the ﬂow problem under consideration is a two-parameter problem with the two small perturbation parameters 1 and 2 . By using the dimensionless quantities, the basic equations for the core region in the cylindrical coordinate system are as follows: ∂u 1 ∂(rv) 1 ∂w + + =0 ∂x r ∂r r ∂θ

(2.12)

∂u w ∂u ∂p ∂ ∂u + ∂u +v + =− + 21 2 γνt u ∂x ∂r r ∂θ ∂x ∂x ∂x ! ∂u 1 ∂u 1 ∂ 2 1 ∂ + + 2 ∂v 2 ∂w + 1 + 1 rγνt + γνt + 1 r ∂r ∂r ∂x r ∂θ r ∂θ ∂x (2.13) " ∂v w ∂v w2 1 ∂p 2 ∂ ∂v ∂u ∂v +v + − =− 2 + + 21 γνt+ u ∂x ∂r r ∂θ r 1 ∂r 1 ∂x ∂r ∂x # ∂ w 1 ∂v 1 ∂ 2γνt+ 1 ∂w v 2 ∂ + ∂v + + + rγνt + γνt r − + r ∂r ∂r r ∂θ ∂r r r ∂θ r r ∂θ r (2.14) " 1 ∂u ∂w w ∂w vw 1 1 ∂p 2 ∂ ∂w + 2 ∂w +v + + =− 2 + + 1 γνt u ∂x ∂r r ∂θ r 1 r ∂θ 1 ∂x r ∂θ ∂x # ∂ w 1 ∂v 1 ∂w v 2 ∂ 1 ∂ + + . r2 γνt+ r + γνt+ + 2 r ∂r ∂r r r ∂θ r ∂θ r ∂θ r (2.15) Here the solution of the system of equations is of interest for the double limit 1 → 0 and 2 → 0. The momentum equation in x-direction degenerates except for the case that the two parameters are coupled such that 1 = 2 = .

(2.16)

This coupling of the viscous effects (2 ) and the geometry (1 ) is typical for the so-called slender-channel theory; see Schlichting and Gersten (2000), page 552.

28

2 Decay of Disturbances in Turbulent Pipe Flow

The dimensionless axial coordinate is now built with the square root of the friction factor λ of the fully developed pipe ﬂow λ x∗ x∗ . (2.17) x= ∗ = R 8 R∗ Hence, the dimensionless velocity components v and w are also connected with λ 1 w∗ 8 v∗ 8 w∗ 1 v∗ = , w = = . (2.18) v= u∗m λ u∗m u∗m λ u∗m When terms of the order O 2 are neglected, (2.12) and (2.13) yield the following equations for the leading order solutions: ∂u 1 ∂(rv) 1 ∂w + + =0 (2.19) ∂x r ∂r r ∂θ ∂u ∂u w ∂u ∂p 1 ∂ 1 ∂ γνt+ ∂u ∂u +v + =− + rγνt+ + . (2.20) u ∂x ∂r r ∂θ ∂x r ∂r ∂r r ∂θ r ∂θ The equations (2.14) and (2.15) are reduced accordingly whereby the terms 2 ∂v/∂x in (2.14) and 2 ∂w/∂x in (2.15) are neglected. Eliminating the pressure in (2.14) and (2.15) leads to the vorticity transport equation for the vorticity component 1 ∂(rw) ∂v ω ∗ R∗ − . (2.21) ωx = x ∗ = um r ∂r ∂θ This transport equation has the form: ∂ωx w ∂ωx ∂u ∂u ∂w 1 ∂u ∂v ∂ωx +v + − ωx + − = ∂x ∂r r ∂θ ∂x ∂r ∂x r ∂θ ∂x ! 1 ∂w v ∂ w 1 ∂v ∂ 1 ∂ 1 ∂ + 2 + + + γνt r r +2 γνt r ∂r r ∂r ∂r r r ∂θ ∂θ r ∂θ r (2.22) ∂v ∂ w 1 ∂v 1 ∂ 1 ∂ 2 ∂ + γνt+ r + γνt+ r − r ∂θ r ∂r ∂r r ∂θ ∂r r r ∂θ ! 1 ∂ ∂u ∂(γνt+ ) ∂u ∂(γνt+ ) 2γνt+ 1 ∂w v + − + . − r r ∂θ r r ∂x ∂θ ∂r ∂r ∂θ u

As it will turn out later the three equations (2.19), (2.20) and (2.22) are sufﬁcient to determine the three velocity components u, v and w. 2.2.3 Boundary Conditions As mentioned in Sect. 2.2.1, the boundary conditions for the core region solution follow from matching this solution with the solution in the viscous wall layer. In the overlap layer the well-known logarithmic velocity distribution occurs.

2.3 Fully Developed Pipe Flow

29

When the resultant velocity in the wall layer is denoted as V ∗ with the components u∗ = V ∗ cos βw and w∗ = V ∗ sin βw , the following matching condition results u∗τ ∂V ∗ (2.23) =− lim ∗ ∗ ∗ ∗ r →R ∂r κ(R − r∗ ) where u∗τ was deﬁned in (2.8) and κ is the Kármán constant. Using dimensionless quantities u∗ V∗ uτ = ∗τ = γ (2.24) V = ∗ , um um the boundary conditions for the functions u, v and w follow from (2.23): lim

r→1

γ cos βw ∂u =− ∂r κ(1 − r) lim v = 0

r→1

lim

r→1

γ sin βw γ sin βw ∂w =− =− ∂r κ(1 − r) κ(1 − r)

(2.25) (2.26) (2.27)

where βw (x, θ) is the constant angle between the resultant velocity and the axial velocity component in the wall layer. Due to the constant resultant shear stress in the wall layer it follows lim νt∗

r ∗ →R∗

∂V ∗ = −u∗τ 2 ∂r∗

(2.28)

and according to (2.7) and (2.23) lim νt+ = κ(1 − r) .

r→1

(2.29)

2.3 Fully Developed Pipe Flow 2.3.1 Velocity Distribution in the Core Region In the fully developed ﬂow (index ∞) u∞ (r) is independent of x and θ. Because of w = 0, βw = 0 and γ = 1 it follows from (2.19) v = 0 and from (2.20) 1 d du∞ dp∞ = rνt+ . (2.30) dx r dr dr Multiplying by rdr and integrating over the cross section lead to 1 dp∞ + du∞ = lim rνt = − , r→1 2 dx dr where (2.25) and (2.29) are taken into account.

(2.31)

30

2 Decay of Disturbances in Turbulent Pipe Flow

The limiting solution for the pipe ﬂow at inﬁnite Reynolds number is the homogeneous ﬂow with the constant velocity u∗m . The solution for large but ﬁnite Reynolds numbers is considered a small perturbation of this limiting solution, where is the perturbation parameter, cf. Schlichting and Gersten (2000). Hence u ∞ (r) has the following asymptotic expansion: u∞ (r) + ... u∞ (r) = 1 +

(2.32)

Combining (2.30), (2.31) and (2.32) ends up with the differential equation for u ∞ (r) d u∞ d rνt+ = −2r . (2.33) dr dr Integration yields r d u∞ =− + . dr νt (r)

(2.34)

If the function νt+ (r) is known, (2.34) can be integrated leading to the so-called defect law r 1 d u∞ dr = F (r) + ln(1 − r) , (2.35) ∞c = u ∞ (r) − u dr κ 0

where the function

r F (r) = 0

1 d u∞ + dr dr κ(1 − r)

(2.36)

is not singular for r → 1. It is r F (0) = 0 ,

F (1) = lim

r→1 0

1 d u∞ + dr = −C . dr κ(1 − r)

It follows from (2.6), (2.7) and (2.32) r lim u ∞ (r)rdr = 0 r→1

(2.37)

(2.38)

0

which determines u ∞c = u ∞ (0) in (2.35): u ∞c

3 −2 = 2κ

1 F (r)rdr = −C .

(2.39)

0

The ﬁnal form of the velocity distribution results from (2.35) as 1 u ∞ (r) = F (r) + ln(1 − r) − C κ which yields in the overlap layer ∞ (r) = lim u

r→1

1 ln(1 − r) − C − C . κ

(2.40)

(2.41)

2.3 Fully Developed Pipe Flow

31

2.3.2 Friction Law ∞ (r) The velocity u∗∞ based on the skin friction velocity u∗τ ∞ is connected with u by u∗ u∗∞ = ∗m + u ∞ . (2.42) u+ ∞ = ∗ uτ ∞ uτ ∞ Matching of u+ ∞ in the overlap layer leads to lim u+ ∞ (r) =

r→1

u∗m + lim u ∞ (r) = +lim u∗τ ∞ r→1 y →∞

1 ln y + + C + κ

(2.43)

where

R∗ u∗τ ∞ (R∗ − r∗ )u∗τ ∞ = (1 − r) (2.44) ∗ ν ν∗ is the wall-layer coordinate and C + a constant; see Schlichting and Gersten (2000). Using (2.41), (2.43) and (2.44) yields ﬁnally the friction law y+ =

1 R∗ u∗τ ∞ u∗m + C+ + C + C . = ln ∗ uτ ∞ κ ν∗

(2.45)

The friction law is frequently considered to be the dependence of the friction factor λ=

8u∗2 τ∞ u∗2 m

(2.46)

on the Reynolds number Re =

u∗m 2R∗ u∗m D∗ = . ν∗ ν∗

(2.47)

From so-called Superpipe Experiments Zagarola and Smits (1998) derived a new friction law which is particularly valid for high Reynolds numbers √ 1 √ = 1.869 log Re λ − 0.241 . (2.48) λ This new friction law is identical with (2.45), when the constants are taken as follows: κ= C

+

= 6.15 (smooth) ,

ln 10 √ = 0.4356 1.869 8 C = 1.51 ,

(2.49) C = −4.36 .

All four constants have independently been determined from the Superpipe Experiments. Note 2.1 (Revised Constants Derived from Superpipe Experiments) When the calculations had been ﬁnished and brought out the results presented in this chapter new and more accurate values of the Kármán constant κ and the additional constants became known, see

32

2 Decay of Disturbances in Turbulent Pipe Flow

McKeon et al. (2003). These improvements of the constants were possible due to corrections of the static pressure measurement error and the velocity gradient Pitot displacement error. The new constants are discussed in Chap. 1 of this book. They are: κ = 0.421;

C + = 5.60;

C = 1.23;

C = −4.28

The calculation have not been repeated by using these new constants, because the authors are convinced, that the results would change only slightly as will be demonstrated later on.

2.3.3 Eddy-Viscosity Distribution To be in accordance with the friction law, the eddy-viscosity distribution ν t+ (r) must satisfy the conditions (2.29), (2.37) and (2.39) with the constants κ, C and C speciﬁed in (2.49). The following distribution satisﬁes all three conditions: κ νt+ (r) or

=

2 α β + + . 2 2 1−r 1 + ar 1 + br2

(2.50)

(1 − r2 )(1 + ar2 )(1 + br2 ) νt+ (r) = κ 2(1 + a)(1 + b)

(2.51)

2b2 (1 + a) . b−a

(2.52)

with α=

2a2 (1 + b) a−b

β=

The constants are α = −0.1411 ,

a = −0.2517

β = 9.0919

b = 6.3171 .

,

(2.53)

In Fig. 2.2 the eddy-viscosity distribution νt+ (r) is shown. The same distribution, but with the new constants according to McKeon et al. (2003) is also shown for comparison. Integration of (2.34) can be carried out easily because of (2.51), which leads to the velocity defect distribution: α β 1 2 2 2 ln(1 + ar ) − ln(1 + br ) − C . (2.54) ln(1 − r ) − u ∞ (r) = κ 2a 2b Its representation in Fig. 2.3 shows also the asymptote for r → 1 according to (2.41). Again the distribution calculated by using the new constants after McKeon et al. (2003) is shown. The deviation is hardly visible. In principle, the eddy-viscosity distribution should be determined by using a turbulence model. However, one of the most popular two-equation turbulence models, the k--model, produces a solution that does not satisfy Zagarola’s friction law, see Gersten and Herwig (1992), p. 534. It seems that even by adjusting the model constants it is not possible to obtain a solution in accordance with Zagarola’s friction

2.3 Fully Developed Pipe Flow

33

0.08

Q+t 0.06

0.04

0.02

0

0

0.2

0.4

0.6

0.8

r

1

Fig. 2.2. Eddy-viscosity distribution of fully developed pipe ﬂow according to (2.51) and (2.53). Asymptote for r → 1 according to (2.29). (——) constants after Zagarola and Smits (1998). (- - - -) constants after McKeon et al. (2003).

law. The k--model is obviously not able to produce the typical distribution of the eddy viscosity known from experiments. The distribution of the eddy viscosity is described quite well by (2.50) which is characterized by a minimum on the pipe axis and a maximum at about r = 0.6. In this simple case of fully developed pipe ﬂow, other two-equation turbulence models and even Reynolds-stress models lead more or less to the same form of a differential equation for νt+ (r). It can therefore be concluded that (2.50) is presently the best available description of the eddy viscosity in fully developed pipe ﬂow. 6

* - u* u m u * 2

0

-2

-4

-6

0

0.2

0.4

0.6

0.8

r

1

Fig. 2.3. Velocity defect distribution of fully developed pipe ﬂow according to (2.54). Asymptote for r → 1 according to (2.41). (——) constants after Zagarola and Smits (1998); (- - - -) constants after McKeon et al. (2003).

34

2 Decay of Disturbances in Turbulent Pipe Flow

In the following this distribution for νt+ (r) will also be applied to pipe ﬂows which deviate from the fully developed state. This is justiﬁed as long as only small deviations are considered.

2.4 Eigenvalue Problem 2.4.1 Basic Equations The general solution of the system of equations (2.19), (2.20) and (2.22) is assumed to be a small perturbation of the fully developed ﬂow. It has the form u(x, r, θ) = u∞ (r) + U (r)e−Λx [A cos kθ + B sin kθ]

(2.55)

v(x, r, θ) =

V (r)e−Λx [A cos kθ + B sin kθ]

(2.56)

w(x, r, θ) =

W (r)e−Λx [A sin kθ − B cos kθ]

(2.57)

dp∞ dP −Λx A + e dx dx

∂p (x) ∂x

=

γ(x, θ)

= 1 + cU e−Λx [A cos kθ + B sin kθ] .

(2.58) (2.59)

Here, A and B are small numbers. Equation (2.59) says that the circumferential component of the wall shear stress has been neglected because of O A2 , O B 2 and O (AB), cf. also (2.68). Since U , V , W and dP/dx are determined apart from an arbitrary factor, cU is used to ﬁx the solution as will be shown later. these functions into (2.19), (2.20) and (2.22) and neglecting terms of Inserting O A2 , O B 2 and O (AB) result in the following three linear ordinary differential equations: (2.60) (rV ) + kW − ΛrU = 0 2 + dP dp∞ k + + rδ0k νt − Λu∞ rU − ru∞ V = −cU r (2.61) νt rU − 2 r dx dx # " k + νt+ 2 1 + 3 W ν r ν rV − 2 k W + kV − r t r r t r W ν+ 2k + 2 + νt rV + k νt − t2 k 3 V + 2k 2 W + 2kV + r r r $ % +Λ u∞ (rW ) + kV + u∞ rW + kU νt+ − cU ku∞ νt+ = 0 .

(2.62)

In (2.61) δ0k is the Kronecker delta (= 1 if k = 0, = 0 if k = 0). The primes refer to differentiation with respect to r.

2.4 Eigenvalue Problem

35

The boundary conditions are: U = 0 , V = 0 ,

r=0: k=0: k=1:

V = 0 , W = 0

U =0,

k = 2, 3 . . . : U = 0 , " r → 1 : U = cU

W =0

V =0,

(2.63)

W =0

1 1 ln(1 − r) − C − C + 1+ κ κ

#

V =0

(2.64) "

W = cW

1 ln(1 − r) − C − C 1+ κ

# .

The boundary conditions at r = 0 have been derived by expanding the solutions into power series of r. The boundary conditions for r → 1 follow from matching with the velocities in the viscous wall layer. In the overlap layer it is in analogy to (2.43) lim

r→1

1 R ∗ u∗ 1 V∗ = ln ∗ τ + ln(1 − r) + C + ∗ uτ κ ν κ =

1 1 1 R∗ u∗τ ∞ ln + ln(1 − r) + C + + ln γ ∗ κ ν κ κ

=

1 1 1 − C − C + ln(1 − r) + ln γ, κ κ

(2.65)

which leads to the boundary conditions (2.64) for U (r) and W (r) for r → 1. For axisymmetric ﬂow (k = 0), integration of (2.60) over the cross section yields r lim

U (r)rdr = 0 ,

r→1

(2.66)

0

i.e., the disturbance velocity distribution has no effect on the volume ﬂux. Integrating (2.61) for k = 0 gives the ﬁnal formula of the disturbance pressure gradient: r 1 dP = −4cU + 4Λ lim u ∞ U rdr . (2.67) r→1 dx 0

Because of this relation between dP/dx and U (r), equations (2.60) to (2.62) can be considered as a system of three ordinary differential equations for the unknown velocity components U (r), V (r) and W (r).

36

2 Decay of Disturbances in Turbulent Pipe Flow

The wall shear stress follows from the velocity gradient at the wall. Since the shear stress in the wall layer is constant and the velocity distribution untwisted, the components of the wall shear stress can also be determined by the velocity gradient in the overlap layer. The formulae for the wall shear stress components are ∗ τwx = γ 2 = 1 + 2cU e−Λx [A cos kθ + B sin kθ] ∗ τw∞

(2.68)

∗ tan βw 1 τwθ = cW e−Λx [A sin kθ − B cos kθ] . = ∗ τw∞

(2.69)

The system of equations (2.61) to (2.62) with the boundary conditions (2.63) and (2.64) is an eigenvalue problem with an inﬁnite number of eigenvalues for each parameter k = 0, 1, 2 . . . . Hence, the general solution of the system is a linear superposition of all eigensolutions. The eigensolutions depend on because u∞ (r) according to (2.32) and the boundary conditions (2.64) are linear functions of . 2.4.2 Limiting Solution for = 0 In the limit = 0 the system (2.60) to (2.64) is simpliﬁed considerably. In (2.61) the terms on the right hand side vanish because of (2.31) and (2.67). The same is true for the term proportional to V . The remaining equation contains only the unknown function U (r) and can be used to determine the eigenvalues Λ. To simplify further it can be set cU = 1. When Λ and U (r) are found, V (r), W (r) and the factor cW can be determined via (2.60), (2.62) and the corresponding boundary conditions. There exists another type of eigensolutions W (r) and V (r), when U (r) vanishes (cU = 0). Then it can be set cW = 1. These eigensolutions describe two-dimensional secondary ﬂows in a pipe. Therefore, two different types of eigensolutions can be distinguished which are independent of each other: the U -type (cU = 1) and the W -type (cU = 0, cW = 1). It can be shown that in both cases the eigensolutions are sets of orthogonal functions with respect to the weight function r u∞ . 2.4.3 Numerical Solutions As already mentioned the function u∞ in (2.61) and (2.62) and the boundary conditions (2.64) depend linearly on . This suggests a splitting of the solution in the form Λ = Λ + Λ ,

, V = V + V , W = W + W &. U = U + U (2.70) When quadratic terms O 2 are neglected, the splitting leads for each eigensolution type and for each k to two eigenvalue problems independent of as is expected in an asymptotic analysis. The numerical solution was carried out by iteration. The eigenvalues had to be estimated such that all boundary conditions were satisﬁed.

2.4 Eigenvalue Problem

37

Because of the singularities in the differential equations for r = 0 and r → 1 and in the boundary conditions for r → 1 the solutions were approximated by series expansions near the ends of the interval. Near the axis power series have been used whereas for r → 1 the series consisted of powers (1−r)n as well as (1−r)n ln(1−r). In Tables 2.1 and 2.2 the main results of the calculation for the ﬁrst three eigenvalues are summarized. The velocity components of the ﬁrst three eigensolutions are shown for Re = 106 ( = 0.0385) and Re = ∞ ( = 0) in Fig. 2.4. It is clear that the eigensolutions will not change much for Reynolds numbers higher than Re = 10 6 . Each set of eigensolutions represents a particular ﬂow structure characterized by the secondary ﬂow described by the velocity components V (r) and W (r). When only the ﬁrst three k-values (k = 0, 1, 2) are considered, the following six ﬂow structures can be identiﬁed: U -type: ring vortex, source-sink pair, source-sink quadrupole (two sources and two sinks). W -type: single longitudinal vortex, longitudinal vortex pair, longitudinal vortex quadrupole. All these ﬂow structures have their individual decay law which is described by as the pipe diameter. The decay an exponential function exp (−αx∗ /D∗ ) with D∗ ' rate α is for all ﬂow structures proportional to = λ/8: Λ √ α = 2Λ = √ λ . 2

(2.71)

There is only one exception. As can be seen from Table 2.2, the eigenvalue of the W type for k = 0 is one order of magnitude smaller than all the others, i.e., Λ W01 = 4. In this special case the decay rate is identical with the friction factor, α = λ. If the length x∗ , within which a value is decayed to one tenth of its value is called ∗ L0.1 , then it follows 2.305 L∗0.1 . (2.72) = D∗ α This tenth-value length is given in Table 2.3 for the various ﬂow structures in descending order. The single longitudinal vortex has the smallest decay rate followed by the source-sink pair. Any given function can be expanded in terms of the set of eigenfunctions. The velocity distributions in the cross section x = 0 have the form u(0, r, θ) = u∞ (r) + UUki (r) [AUki cos kθ + BUki sin kθ] (2.73) k=0,... i=1,...

w(0, r, θ) =

WWki (r) [AWki sin kθ − BWki cos kθ]

(2.74)

k=0,... i=1,...

where the subscripts U and W indicate the eigensolution type and i the number of the eigensolution in a particular set.

38

2 Decay of Disturbances in Turbulent Pipe Flow

U-type 1

U(r) U(0)

W (r) W (0) i=1

0.5

k=0

W-type

1

i=1

0.5

i=2

0

0

i=3

i=3 i=2

- 0.5

-1

0

0.25

0.5

- 0.5

0.75

r

1

0.75

-1

0

0.25

W(0)

0

-1

i=3 0.25

0.5

0.75

r

1

-1.5

0

0.24

1

U(r) U (0)

W (r)

W (0)

0.16

0.5

1

0.75

r

1

i=3

0.25

i=1 0

i=3

i=2

0

0.5

0.75

i=2

- 0.5

i=3

0.25

0.5

i=1

0.08

0

r

i=2

- 0.5

i=2

- 0.08

0.75

i=1

0.5 0

k=2

1

i=1

0.25

0

r

1

0.5

-0.25

0.75

W(r)

U(r) U (0)

k=1

0.5

r

1

-1

0

0.25

0.5

Fig. 2.4. Velocity components of the ﬁrst three eigensolutions for k = 0, 1 and 2; (—–) Re = 106 ( = 0.0385); (- - - -) Re = ∞ ( = 0).

ki − W ki U ki Table 2.1. Characteristic results of the eigenvalue problem (U -type). After deﬁnition in (2.64): cW = (1 + /κ)W ki + lim W r→1

k=0 2 3 i=1 0.9555 2.6761 5.2860 Λ1i -2.3400 2.3400 -2.3400 dU 1i /dr r=0

Λ0i U 0i (0)

1

1

2

U 0i rdr

2

0i (0) U

1 0

1

0.0444 -0.4349 0.5665

W 1i (1)

0.1342 0.1365 0.0327 W 2i (1)

0i rdr 2.3658 -1.0432 -1.6216 2 U 0i U 2

u ∞ U 0i rdr

1

0

1

1

1i rdr U 1i U

-0.7349 -1.9716 -1.8131 2 0

1

2

0.1493 0.1416 0.0250

1.1868 0.4817 0.9891

4.5194 5.2667 7.0681 -55.952 1299.5 -7361.8

r=0

2i rdr U 2i U

-0.8743 -1.7572 -1.7087

2

0

1.2726 1.4716 -1.4863 r→1

0.2163 -2.0193 3.9938

r=0

u ∞ U 2i rdr

-0.3338 0.1137 0.0708

0

1i (0) W 1i − W 1i U 1i W

3.0332 4.0385 6.0103 Λ2i 2i /dr 2 7.0182 12.481 -45.927 d2 U

u ∞ U 1i rdr

0.3903 0.1144 0.0647

dW 2i /dr

0.2213 0.1899 0.1415

2i /dr dW

-0.4250 -0.0079 0.0336

r=0

2i − W 2i U 2i W

3.6519 3.2103 -19.955 r→1

0.1838 0.6128 1.1215

2.4 Eigenvalue Problem

2

W 1i (0)

r=0

5.0349 -28.125 66.551

2

0

2.1514 5.9031 9.0425 Λ1i 1i /dr dU -15.624 -11.002 15.531

k=2 2 3 i=1 0.5299 2.6161 5.2882

U 2i rdr

0.3297 0.2476 0.1664

0

Λ0i

0

1

U 1i rdr

0.3521 0.2007 0.1429

0

k=1 2 3 i=1 0.1980 1.6871 3.8735 Λ2i 1.8699 -6.6563 10.928 d2 U 2i /dr 2 r=0

39

40

Table 2.2. Characteristic results of the eigenvalue problem (W -type).

Λ0i

dW 0i /dr r=0

1 2 W 0i rdr

3 3.9061

k=1 2 3 i=1 0.5299 2.6300 5.3762 Λ2i

Λ1i

1.000 -5.6174 10.3291 W 1i (0)

1 2 0.2500 0.2409 0.1666 W 1i rdr

0

4.0000 3.2131

3.8305

dW 2i /dr r=0

1 2 0.1553 0.1444 0.1171 W 2i rdr

-0.7101 1.0662 -1.0973

0

Λ0i

k=2 2 3 i=1 0.9467 3.6500 6.8339 -2.0867 4.3573 -7.2139 0.1105 0.1005 0.0892

0

Λ1i

4.7976 3.5246 3.8926 Λ2i

5.3000 3.6197 0.1909

0i /dr 2i /dr 1i (0) dW 4.3586 -12.4343 -10.5386 W -8.1690 1.6878 2.0782 dW -155.344 63.117 -18.553 r=0 r=0 1 1 1

0i rdr -0.8029 -1.5840 -1.6165 2 W 1i W 1i rdr -0.3832 -0.6247 -1.0009 2 W 2i W 2i rdr -0.2788 -0.2080 -0.6166 2 W 0i W 0

1

0

2

u ∞ W 0i rdr 0

1 -0.4012 0.1317

0

2

u ∞ W 1i rdr

0.0926 0

1

2

u ∞ W 2i rdr

-0.3618 -0.0926 -0.0272 0

-0.3374 -0.1518 -0.0739

2 Decay of Disturbances in Turbulent Pipe Flow

k=0 2 i=1 0.0000 1.7149

2.4 Eigenvalue Problem

41

Table 2.3. Tenth-value length according to (2.72) for various ﬂow structures in descending order. Λ

1. ﬂow structure 1 2 3 4 5 6

single longitudinal vortex source-sink pair vortex pair source-sink quadrupole vortex quadrupole ring vortex

0 0.198 0.530 0.530 0.947 0.956

L∗0.1 /D∗ Re = 106

194 151 56 56 32 31

The coefﬁcients are in a ﬁrst approximation determined by 2π ( (1

AU0i =

u∞ (r) [u(0, r, θ) − u∞ (r)] UU0i rdrdθ

0 0

2π

(1

(2.75) 2 rdr u∞ UU0i

0 2π ( (1

AUki =

u∞ (r) [u(0, r, θ) − u∞ (r)] UUki cos(kθ)rdrdθ

0 0

(1

π

k = 1, 2 . . . 2 rdr u∞ UUki

0 2π ( (1

BUki =

(2.76) u∞ (r) [u(0, r, θ) − u∞ (r)] UUki sin(kθ)rdrdθ

0 0

π

k = 1, 2... (2.77)

(1

2 rdr u∞ UUki

0 2π ( (1

BW0i = − 0

u∞ (r)w(0, r, θ)WW0i rdrdθ

0

2π

(1

(2.78) 2 rdr u∞ WW0i

0 2π ( (1

AWki =

u∞ (r)w(0, r, θ)WWki sin(kθ)rdrdθ

0 0

π

(1

k = 1, 2...

(2.79)

2 u∞ WWki rdr

0 2π ( (1

BWki = −

u∞ (r)w(0, r, θ)WWki cos(kθ)rdrdθ

0 0

π

(1 0

k = 1, 2... 2 rdr u∞ WWki

(2.80)

42

2 Decay of Disturbances in Turbulent Pipe Flow

The integrals in the denominators can be determined by superposition of the integrals given in Tables 2.1 and 2.2. The general solutions for u(x, r, θ) and w(x, r, θ) are obtained when in (2.73) and (2.74) the eigensolutions are multiplied by exp (−Λ Uki x) and exp (−ΛWki x), respectively.

2.5 Comparisons with Experimental Results 2.5.1 Axisymmetric Flow without Swirl Experimental investigations of the decay of distorted axisymmetric turbulent pipe ﬂows have been carried out by Laws et al. (1987). The disturbed inlet proﬁles were generated by gauze screens. The test Reynolds number was Re = 2.52 × 10 5 ( = 0.0434). Figure 2.5 shows for a typical example (case 3 from x∗ /D∗ = 8 to x∗ /D∗ = 19.5) the comparison between theory and experiment. The coefﬁcients of the series u(x, r) = u∞ (r) +

3

AU0i UU0i exp −2ΛU0i

i=1

x∗ −8 D∗

(2.81)

are AU01 = 0.0648, AU02 = 0.0012, AU03 = −0.0392. In cross section x∗ /D∗ = 19.5 the series in (2.81), reduced to only the ﬁrst term, is in good agreement with the experimental results. 1.2

1.2

u

u

0.8

0.8

0.6

0.6

x*/D* = 8.0

0.4

0.4

0.2

0.2

0.0 0.0

0.25

0.50

0.75

r

1.0

0.0 0.0

x*/D* = 19.5

0.25

0.50

0.75

r

1.0

Fig. 2.5. Distorted velocity distributions of an axisymmetric turbulent pipe ﬂow without swirl Re = 2.52×105 ( = 0.0434); (◦) experiments by Laws et al. (1987); (—–) theory according to (2.81)

2.5 Comparisons with Experimental Results

43

2.5.2 Axisymmetric Flow with Swirl Analysis In axisymmetric fow with swirl (W -type, k = 0) equation (2.62) can be integrated and reduces to the simple equation WW0i + 3 + ΛW0i r2 u∞ WW0i = 0 . (2.82) νt r r Hence, the disturbances in the circumferential and axial direction are not coupled in this case. It can be shown that the eigensolutions are a set of orthogonal functions with respect to the weight function ru∞ , i.e., ⎧ 1 ⎨ 0 for i = j (2.83) lim ru∞ WW0i WW0j dr = r→1 ⎩ c for i = j 0 The ﬁrst eigensolution can be given in analytical form: ΛW01 = 4

,

WW01 = ru∞ .

(2.84)

Hence, the circumferential velocity distribution far downstream is w(x, r) = −BW01 ru∞ (r) exp [−λx∗ /D∗ ] .

(2.85)

The decay rate α according to (2.71) is in this case identical with the friction factor λ of the fully developed pipe ﬂow at the given Reynolds number. In the limit Re → ∞ (u∞ = 1) the circumferential velocity w(x, r) is proportional to the radius r (rigid body rotation). It is worth mentioning that the coefﬁcient BW01 according to (2.80) is proportional to the commonly used swirl number, cf. Steenbergen (1995) p. 15, 1 S=2 0

u∗ w ∗ 2 r dr = − BW01 ∗ ∗ um um 2

(2.86)

when the Reynolds number is high (WW01 ≈ r) and u(r) is substituted by u∞ (r). In practice this latter condition is usually satisﬁed because the decay rate for the axial disturbances is much higher than that for the circumferential disturbances, see Table 2.3 (cases 1 and 6). Under the conditions mentioned only the ﬁrst eigensolution determines the swirl number. All other eigensolutions have no contribution to S because of the orthogonality. The radial distribution of the swirl angle far downstream (axial and circumferential disturbances of second and higher eigensolutions already decayed) follows from (2.85) tan β(x, r) =r (2.87) tan βw (x) when (2.69) is taken into account.

44

2 Decay of Disturbances in Turbulent Pipe Flow

Comparison with Results of Reader-Harris Reader-Harris (1994) has analyzed the turbulent pipe ﬂow with swirl by a method similar to the special case (W -type, k = 0) presented here. In contrast to the Reynolds number independence of the asymptotic analysis he calculated the eigensolutions individually for each Reynolds number. He found the mean value of the ﬁrst decay rate for the Reynolds numbers 105 ≤ Re ≤ 107 as α1 = 1.07λ in fairly good agreement with α1 = λ in (2.85). An extrapolation of his α-values to Re → ∞ would give α = λ. Reader-Harris compared his results of the decay rate α as function of λ and of the swirl angle distribution according to (2.87) with various experiments and found good agreement.

turbulence model

Reader-Harris

Prandtl

Zagarola

McKeon

constants

κ C + (smooth)

0.40 5.5

0.407 5.0

0.436 6.15

0.421 5.6

3.77

3.01

2.85

3.05

Re = 106

Table 2.4. Model constants and the ﬁrst three decay rates α according to (2.71) for four different turbulence models.

λ α1 α2 α3

0.0115 0.0123 0.172 0.391

0.0116 0.0116 0.145 0.344

0.0119 0.0119 0.144 0.303

0.0118 0.0118 0.141 0.312

− C+C

The decay rates depend slightly on the turbulence model used, as can be seen in Table 2.4. In this table the ﬁrst three decay rates are listed for four different turbulence models, the model of Reader-Harris and the model of the present paper applied to the friction law by Prandtl, by Zagarola and by McKeon. The two last models show only very little difference, as mentioned earlier. The type of turbulence model seems to have an increasing effect on higher decay rates. Reader-Harris investigated also the effect of roughness on the decay of swirl. Since C + in (2.45) is the only number affected by the roughness, the whole analysis presented here is the same. Only the value for a given Reynolds number increases due to roughness. Increasing roughness is equivalent to decreasing the Reynolds number at smooth surface condition. Roughness increases the decay rates in complete agreement with the results by Reader-Harris and with further investigations on roughness effects by Senoo and Nagata (1972) and Morrison and Tung (1999). 2.5.3 Three-Dimensional Flow Experimental data which are accurate and detailed enough in the whole ﬁeld of the pipe ﬂow with small disturbances for the purpose of comparison with the theory are not known to the authors.

2.5 Comparisons with Experimental Results

45

0.06

AU01

a

0.04

0.02

0 10

20

30

40

x*/ D*

50

0.04

b

CU11 CU21

0.03

0.02

0.01

0 10

20

30

40

x*/ D*

50

0.08

H .CW11 H .CW21

0.06

c

0.04

0.02

0 10

20

30

40

x*/ D*

50

Fig. 2.6. Decay of the coefﬁcients (2.73), (2.74) and (2.88) downstream of a bend. Re = 2.2 × 105 ( = 0.0439). (◦) () experiments by Wildemann (2000); (——) ∼ exp [−α (x∗ /D∗ − 11)] adjusted to experiment at x∗ /D∗ = 11, α according to (2.71), Λ from Tables 2.1 and 2.2. (- - - -) ∼ exp [−α (x∗ /D∗ − 21)] adjusted to experiment at x∗ /D∗ = 21

46

2 Decay of Disturbances in Turbulent Pipe Flow

In technical practice the decay process is quite often already so close to the fully developed state that already the ﬁrst eigensolutions are sufﬁcient to describe the ﬂow. In this case the coefﬁcients in (2.73) and (2.74) can be determined easily as Gersten et al. (2001) and Wildemann et al. (2002) have shown. The authors measured the circumferential distributions of the wall shear stress components and could determine the coefﬁcients by using (2.68) and (2.69). Figure 2.6 shows as a typical example the decay of the coefﬁcients in the pipe downstream of a bend. For k = 1 and k = 2 the coefﬁcients Ak1 and Bk1 have been combined to 2 (2.88) Ck1 = A2k1 + Bk1 characterizing the intensity of the corresponding ﬂow structure. The coefﬁcient AW01 could not be determined since the swirl in the experiment was very small and showed no decaying tendency. If the data points lie exactly on an exponential curve, the ﬁrst eigensolution is sufﬁcient to describe the decay process. This is the case for AU01 , CU21 and CW21 where the agreement between theory and experiment is very good. On the other hand, the ﬁrst eigensolution is not sufﬁcient to describe the decay process for C U11 and this is also true, though to a lesser degree, for CW11 . The value CU11 represents the structure of a source-sink pair. This structure has the smallest decay rate of all structures present in this ﬂow according to Table 2.3. The dotted line in Fig. 2.6b, adjusted to the experiments at x∗ /D∗ = 21, shows excellent agreement between theory and experiment. The same may be true for CW11 in Fig. 2.6c, representing the vortex-pair structure. From this comparison between theory and experiment it can be concluded that the decay process downstream of the bend under consideration is purely exponential at x∗ > 11D∗ for AU01 , CU21 , CW21 (ring vortex, source-sink quadrupole, vortex quadrupole) and at about x∗ > 20D∗ for CU11 and CW11 (source-sink pair, vortex pair). The terms for k ≥ 3 are in practice negligible as could also be seen from the experimental data.

References Barbin AR, Jones JB (1963) Turbulent ﬂow in the inlet region of a smooth pipe. J. Basic Eng. 85:29–34 Cebeci T, Chang KC (1978) A general method for calculating momentum and heat transfer in laminar and turbulent duct ﬂows. Numerical Heat Transfer 1:39–68 Gersten K, Herwig H (1992) Strömungsmechanik. Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig/Wiesbaden Gersten K, Klika M (1998) The decay of three-dimensional deviations from the fully developed state in laminar pipe ﬂow. In: Rath HJ, Egbers Ch. (Eds.): Advances in Fluid Mechanics and Turbomachinery, Springer-Verlag, Berlin/Heidelberg, pp. 17–28 Gersten K, Merzkirch W, Wildemann C (2001) Verfahren und Vorrichtung zur Korrektur fehlerhafter Messwerte von Durchﬂussmessgeräten infolge gestörter Zuströmung. Patent No. 197 24 116

References

47

Herwig H, Voigt M (1995) Turbulent entrance ﬂow in a channel: an asymptotic approach. In: Bois PA et al. (Eds.):Asymptotic Modelling in Fluid Mechanics, Springer-Verlag, Berlin/Heidelberg/New York, pp. 51–58 Kalkühler K (1998) Experimente zur Entwicklung der Geschwindigkeitsproﬁle und Turbulenzgrößen hinter verschiedenen Gleichrichtern. Fortschritt-Berichte, Reihe 7, Nr. 339, VDI-Verlag, Düsseldorf Kitoh O (1991) Experimental study of turbulent swirling ﬂow in a straight pipe. J. Fluid Mech. 225:445–479 Klein A (1981) REVIEW: Turbulent developing pipe ﬂow. J. Fluids Eng. 103:243–249 Laws EM, Lim EH, Livesey JL (1987) Momentum balance in highly distorted turbulent pipe ﬂow. Exp. in Fluids 5:36–42 McKeon BJ, Morrison JF, Jiang W, Li J, Smits AJ (2003) Revised log-law constants for fullydeveloped turbulent pipe ﬂow. In: AJ Smits (Ed.): IUTAM Symposium on Reynolds Number Scaling in Turbulent Flow. Kluwer Academic Publishers, Dordrecht Mickan B (1999): Systematische Analyse von Installationseffekten sowie der Efﬁzienz von Strömungsgleichrichtern in der Großgasmengenmessung. Doctor thesis, University Essen Morrison GL, Tung K (1999) The effect of pipe wall roughness upon the ﬂow ﬁeld downstream of two close coupled 90◦ out of plane elbows. Proceedings 4th International Symposium on Fluid Flow Measurement, Denver, Colorado, June 27–30, 1999 Norman RS, Mattingly GE, Mc Faddin SE (1989) The decay of swirl in pipes. Proceedings Intern. Gas Research Conference, 312–321 Parchen RR (1993) Decay of swirl in turbulent pipe ﬂows. Ph. D. thesis, Technical University Eindhoven Reader-Harris MJ (1994) The decay of swirl in a pipe. Int. J. Heat and Fluid Flow 15:212–217 Senoo Y, Nagata T (1972) Swirl ﬂow in long pipes with different roughness. Bulletin of the JSME, 15:1514–1521 Schlichting H, Gersten K (2000) Boundary-Layer Theory. Springer-Verlag, Berlin / Heidelberg Steenbergen W (1995) Turbulent pipe ﬂow with swirl. Ph. D. thesis, Technical University Eindhoven Steenbergen W, Voskamp J (1998) The rate of decay of swirl in turbulent pipe ﬂow. Flow Meas. Instrum. 9:67–78 Sudo K, Sumida M, Hibara H (1998) Experimental investigation of turbulent ﬂow in a circularsectioned 90-degree bend. Exp. in Fluids 25:42–49 Voigt M (1995) Die Entwicklung von Geschwindigkeits- und Temperaturfeldern in laminaren und turbulenten Kanal- und Rohrströmungen aus asymptotischer Sicht. VDI-FortschrittBerichte, Reihe 7, Nr. 262, VDI-Verlag, Düsseldorf Wildemann C (2000) Ein System zur automatischen Korrektur der Messabweichung von Durchﬂussmessgeräten bei gestörter Anströmung. Doctor thesis, Universität Essen. See also: Fortschritt-Berichte VDI, Reihe 8, Nr. 868 Wildemann C, Merzkirch W, Gersten K (2002) A universal, nonintrusive method for correcting the reading of a ﬂow meter in pipe ﬂow disturbed by installation effects. J. of Fluids Eng. 124:650-656 Zagarola MV, Smits AJ (1998) Mean-ﬂow scaling of turbulent pipe ﬂow. J. Fluid Mech. 373, 33-79

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow Klaus Gersten, Heinz-Dieter Papenfuss Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

A set of optimal characteristic parameters for turbulent pipe ﬂows is proposed to quantify the disturbances, i.e., the deviations from the fully developed ﬂow. These optimal characteristic parameters are based on the eigensolutions of the threedimensional ﬂow ﬁeld in the pipe and have the following features: Any desired accuracy can be achieved. The number of parameters is a minimum for a given accuracy. Each parameter is independent of the number of parameters being used. The ﬂow ﬁeld can be reconstructed from the characteristic parameters. The characteristic parameters can be easily determined by a nonintrusive experiment. As an example, the ﬂow ﬁeld approximately ten diameters downstream of a single bend at a Reynolds number of Re ≈ 2 × 105 is analyzed.

3.1 Introduction Flows in straight pipes are often not fully developed. Deviations from the fully developed state may occur due to bends, valves or other pipe ﬁttings. Even swirl can be found downstream of a pair of bends installed in perpendicular planes. The knowledge of these deviations from the fully developed condition is of great importance. For example, these deviations can cause signiﬁcant errors in the measurement of the mass ﬂow rate. In order to quantify the intensity of various ﬂow disturbances in the pipe certain characteristic parameters are used. These parameters are sometimes denoted as “ﬂow indices”. The axisymmetric ﬂow without swirl can be characterized by the blockage factor, which is essentially the ratio between the maximum velocity and the average velocity, cf. Klein (1981) and Cebeci and Chang (1978). The swirl intensity can be determined by the swirl number, which is a dimensionless radial moment of the circumferential momentum ﬂux, see Steenbergen and Voskamp (1998). Corresponding characteristic parameters have been deﬁned for three-dimensional disturbances by Morrison et al. (1992, 1995), Sudo et al. (1998) and Mickan (1999). All these characteristic parameters mentioned have been applied only to special cases and are not

50

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow

sufﬁcient to describe properly and accurately enough the general three-dimensional ﬂow ﬁeld.

3.2 Objective In the following a set of characteristic parameters will be proposed to describe the general three-dimensional turbulent pipe ﬂow. The proposal has the following properties: 1. The number of characteristic parameters can be as high as necessary to guarantee the desired accuracy. 2. The number of parameters is a minimum for a given accuracy in comparison with other possible sets of parameters. 3. Each parameter is independent of the total number of parameters used. 4. From a given set of characteristic parameters the ﬂow ﬁeld can be reconstructed (“inversion”). 5. The characteristic parameters can be easily determined by experiment. Because of these properties the designation “optimal characteristic parameters” has been chosen.

3.3 Formulae for the Optimal Characteristic Parameters The decay of three-dimensional deviations from the fully developed state in turbulent pipe ﬂow far downstream at high Reynolds numbers has be investigated by Gersten and Papenfuss in Chap. 2 of this book. The general solution has been expanded into series of eigensolutions. These are for the axial velocity component u(x, r, θ) = u∞ (r) + UUki (r)e−ΛUki x [AUki cos kθ + BUki sin kθ] k=0,... i=1,...

(3.1) and for the azimuthal velocity component w(x, r, θ) = WWki (r)e−ΛWki x [AWki sin kθ − BWki cos kθ] . (3.2) k=0,... i=1,...

A cylindrical coordinate system has been used according to Fig. 3.1. It should be noted, that all velocity components are based on the average velocity u m , in contrast to Chap. 2 of this book, where the radial and the circumferencial velocity components are based on um . The velocity components are double series. In circumferential direction it is a Fourier series with k = 0, 1, 2, . . . as running parameter. The decay in downstream direction is given by a linear combination of exponential functions. The eigensolutions UUki (r) and WWki (r) have been documented in detail by Gersten and Papenfuss. They have been normalized in a way that they have the same wall shear

3.3 Formulae for the Optimal Characteristic Parameters

51

R

r,v

x,u

T,w

Fig. 3.1. Geometry and coordinate system. The radial coordinate r is normalized by the pipe radius R, the velocity components u, v, w by the average velocity um .

stress as the fully developed ﬂow at the same Reynolds number. The eigensolutions do not contribute to the ﬂow rate and depend only slightly on the Reynolds number. The characteristic parameters are given by: 2π ( (1

AU0i =

u∞ (r) u(0, r, θ) − u∞ (r) UU0i rdrdθ

0 0

2π

(1

(3.3) 2 rdr u∞ UU0i

0 2π ( (1

AUki =

u∞ (r) u(0, r, θ) − u∞ (r) UUki cos(kθ)rdrdθ

0 0

π

(1 0

2π ( (1

BUki =

(3.4) 2 rdr u∞ UUki

u∞ (r) u(0, r, θ) − u∞ (r) UUki sin(kθ)rdrdθ

0 0

π

(1 0

2π ( (1

BW0i = − 0

(3.5) 2 rdr u∞ UUki

u∞ (r)w(0, r, θ)WW0i rdrdθ

0

2π

(1 0

(3.6) 2 rdr u∞ WW0i

52

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow 2π ( (1

AWki =

u∞ (r)w(0, r, θ)WWki sin(kθ)rdrdθ

0 0

π

(1

(3.7) 2 rdr u∞ WWki

0 2π ( (1

BWki = −

u∞ (r)w(0, r, θ)WWki cos(kθ)rdrdθ

0 0

π

(1 0

2 rdr u∞ WWki

(3.8)

(k = 1, 2 . . .)

3.4 Fully Developed Flow The velocity distribution of the fully developed ﬂow is u∞ (r) , u∞ (r) = 1 + ˜

(3.9)

where u ˜∞ (r) can be described by β α 1 2 2 2 ln 1 + ar − ln 1 + br −C . ln 1 − r − u ˜∞ (r) = κ 2a 2b

(3.10)

The constants α und β depend on a and b. Integration of this velocity distribution leads ﬁnally to a friction law which is a correlation between the friction factor 8τw∞ ∆p∞ D = = 82 λ= ρ 2 2 L ρ u m u 2 m

(3.11)

and the Reynolds number Re =

um D . ν

(3.12)

The constants a, b, κ and C in (3.10) have been chosen in such a way that the friction law derived from the Superpipe Experiments by Zagarola and Smits (1998) is reproduced: √ 1 √ = 1.869 log Re λ − 0.241 . λ

(3.13)

The constants are: a = −0.2517 ,

b = 6.3171

α = −0.1411 ,

β = 9.0919

κ = 0.4356

C = −4.36 .

,

(3.14)

3.5 Minimal Program

53

3.5 Minimal Program In practice it is sufﬁcient to use three terms of the Fourier series, i.e., k = 0, 1 and 2. As already mentioned, the series are linear combinations of exponential functions. Experiments, however, often show a pure exponential decay. This means that all higher eigensolutions are already damped out except the leading one. Far downstream a pure exponential decay is justiﬁed. The minimal number of eigensolutions involved is therefore 3 times 2 = 6. Each eigensolution represents a particular ﬂow structure, which has its individual decay law. Table 3.1 gives the ﬁrst six ﬂow structures. Their decay behavior is determined by the tenth-value length L 0.1 , which is the length within which a value is decayed to one tenth of its original value. In Table 3.2 the six eigensolutions are listed. It is Uki (r) , UUki (r) = U Uki (r) + U (3.15) &Wki (r) . WWki (r) = W Wki (r) + W The eigensolutions are sets of orthogonal functions, i.e., ⎫ (1 ⎪ ⎪ u∞ (r)UUki UUkj rdr = 0 ⎪ ⎪ ⎬ 0 for i = j ⎪ (1 ⎪ ⎪ u∞ (r)WWki WWkj rdr = 0 ⎪ ⎭

(3.16)

0

with the consequence that the characteristic parameters according to (3.3) to (3.8) are independent of the number of parameters being used. It is noteworthy that BW0.1 is proportional to the commonly used swirl number S, i.e., BW0.1 = 2S, when u(r) is substituted by the velocity distribution of the fully developed ﬂow u∞ (r). In practice this latter condition is usually satisﬁed because the decay rate for axial disturbances is much higher than that for circumferential disturbances, see Table 3.1. Table 3.1. Tenth-value length L0.1 for the ﬁrst six ﬂow structures in descending order. Re = 106 . ﬂow structure 1 2 3 4 5 6

single longitudinal vortex source-sink pair vortex pair source-sink quadrupole vortex quadrupole ring vortex

L0.1 /D 194 151 56 56 32 31

54

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow

Since the ﬂow structures are usually determined by two characteristic parameters (intensity and orientation) except the axisymmetric ones, ten characteristic parameters are needed to describe the ﬁrst six ﬂow structures listed in Table 3.1. These can be determined using (3.3) to (3.8). However, if the velocities and their ﬁrst and second derivatives at the pipe axis are known, these ten characteristic parameters are given by: (3.17) AU01 = [u(0) − u∞ (0)] /UU01 (0) BU11 = u (0, θ = 90◦ ) /UU11 (0)

(3.18)

(0) AU11 = u (0, θ = 90◦ ) /UU11

(3.19)

(0) (0)AU01 ] /UU21 AU21 = [u (0, θ = 0◦ ) − u∞ (0) − UU01

(3.20)

(0)AU01 ] /UU21 (0) BU21 = [u (0, θ = 45◦ ) − u∞ (0) − UU01

(3.21)

√ , BW01 = − [w (0, θ = 45◦ ) + w (0, θ = 90◦ )] / 2 √ - (0) +w (0, θ = 112.5 ) / 1 + 2 WW01

(3.22)

AW11 = w (0, θ = 90◦ ) /WW11 (0)

(3.23)

BW11 = −w (0, θ = 0◦ ) /WW11 (0)

(3.24)

AW21 = [w (0, θ = 45◦ ) − WW01 (0)BW01 ] /WW21 (0)

(3.25)

BW21 = − [w (0, θ = 90◦ ) − WW01 (0)BW01 ] /WW21 (0)

(3.26)

◦

The values and their derivatives of the eigenfunctions at the axis are given in Table 3.2. Furthermore it is u∞ (0) = 1 + 4.36 ,

u∞ (0) = −25.14 .

(3.27)

Table 3.2. Components of the six eigensolutions UUk0 (r) and WWk0 (r) for k = 0, 1, 2 according to (3.15) after Gersten and Papenfuss in Chap. 2 of this book.

r

U01 U11 U21 W01 U U01 (0) U (0) U U11 (0) U (0) U U21 (0) U (0) W W01 (0) W (0) = 28.101 = 303.73 = 1.8699 = 7.0182 = 5.0349 = -55.952 = 1.0000 = 4.3586 U01 U11 U21 W01 U U01 U U11 U U21 W W01 W W11 U U U W

W11 W

W21 W W21 (0) W (0) = -2.0867 = -155.34 W21 W W21 W

-2.3400

-15.624

0.0000

0.0000

0.0000

0.0000

0.0000

0.0000

-0.7101

-8.1690

0.0000

0.0000

-2.2048

-14.177

0.1836

0.6719

0.0245

0.0906

0.1000

0.4236

-0.6734

-5.0847

-0.1866

-8.6506

0.2

-1.8544

-10.564

0.3494

1.1870

0.0910

0.3128

0.2000

0.7799

-0.5713

-2.8733

-0.3218

-5.3541

0.3

-1.4006

-6.2335

0.4892

1.4695

0.1849

0.5626

0.3000

1.0240

-0.4235

-1.2175

-0.3788

-3.4354

0.4

-0.9343

-2.3003

0.6037

1.5173

0.2931

0.7473

0.4000

1.1324

-0.2475

0.0643

-0.3546

-1.8481

0.5

-0.5017

0.7217

0.6976

1.3521

0.4075

0.8034

0.5000

1.0874

-0.0555

0.9671

-0.2578

-0.4599

0.6

-0.1179

2.6973

0.7759

0.9833

0.5243

0.6796

0.6000

0.8616

0.1460

1.4556

-0.0992

0.6306

0.7

0.2176

3.6197

0.8426

0.3842

0.6420

0.3074

0.7000

0.3980

0.3533

1.4579

0.1118

1.2469

0.8

0.5107

3.4231

0.9008

-0.5523

0.7603

-0.4541

0.8000

-0.4372

0.5650

0.8057

0.3681

1.1011

0.9

0.7691

1.6718

0.9527

-2.2176

0.8794

-2.0397

0.9000

-2.0689

0.7805

-1.0180

0.6651

-0.4868

0.95

0.8876

-0.4488

0.9768

-3.8845

0.3995

-3.7320

0.9500

-3.7638

0.8897

-3.0349

0.8279

-2.5859

0.96

0.9105

-1.1232

0.9815

-4.4146

0.9515

-4.2763

0.9600

-4.3084

0.9117

-3.6713

0.8616

-3.2665

0.97

0.9332

-1.9731

0.9862

-5.1024

0.9636

-4.9833

0.9700

-5.0069

0.9337

-4.4846

0.8956

-4.1384

0.98

0.9557

-3.1297

0.9908

-6.0538

0.9757

-5.9597

0.9800

-5.9837

0.9558

-5.5848

0.9300

-5.3145

0.99

0.9780

-5.0017

0.9954

-7.7231

0.9878

-7.6640

0.9900

-7.6329

0.9779

-7.4419

0.9648

-7.2752

1.00

1.0000

−1/

1.0000

−1/

1.0000

−1/

1.0000

−1/

1.0000

−1/

1.0000

−1/

3.5 Minimal Program

0 0.1

55

56

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow

3.6 Example The turbulent ﬂow in a pipe downstream of a single bend has been measured in detail by Mickan (1999) using Laser-Doppler anemometry. Figure 3.2 shows the velocity distributions ten diameters downstream of the bend. The thirty ﬁrst characteristic parameters have been determined from these experimental data and are listed in Table 3.3. The velocity distributions reconstructed from these characteristic parameters are compared with the experimental data in Fig. 3.2 (solid lines). The dotted lines result from assuming that the ﬁrst six eigensolutions are sufﬁcient to describe the ﬂow ﬁeld (minimal program, k = 0, 1, 2; i = 1; U - and W -type). Table 3.3. The ﬁrst 30 characteristic parameters AUki , BUki , AWki , BWki (k = 0, 1, 2; i = 1, 2, 3) for the pipe ﬂow in the cross section 10 diameters downstream of a bend according to measurements by Mickan (1999). Re = 1.9 × 105 ( = 0.0445). k k=0 k=1 k=2

i

AUki

BUki

AWki

BWki

1 2 3 1 2 3 1 2 3

0.0541 0.0170 -0.0113 -0.0290 -0.0046 0.0140 -0.0115 0.0012 0.0045

-0.0138 -0.0036 -0.0045 0.0060 0.0032 -0.0046

-0.0065 0.0004 0.0080 0.0190 -0.0034 -0.0053

0.0020 0.0027 -0.0024 0.0131 0.0063 -0.0078 0.0086 -0.0025 -0.0045

3.7 Experimental Determination of the Characteristic Parameters (Minimal Program) In case that the ﬁrst eigensolutions are sufﬁcient to describe the ﬂow, the characteristic parameters can be determined quite easily from experiments as Gersten et al. (2001) and Wildemann (2000) have shown, see also Chap. 14 of this book. The authors measured the circumferential distributions of the wall shear stress components and could determine the coefﬁcients by using the formulas derived from (3.1) and (3.2): τwx =1+2 [AUki cos kθ + BUki sin kθ] (3.28) τw∞ k=0,...

τwθ = [AWki sin kθ − BWki cos kθ] . τw∞

(3.29)

k=0,...

Figure 3.3 shows a typical result of the measured wall shear stress distributions from Wildemann (2000). The coefﬁcients of the Fourier series of these distributions are

3.7 Experimental Determination of the Characteristic Parameters (Minimal Program)

57

1.2

u 1.0 0.9 0.8

T = 180°

T = 225°

T = 0°

T = 45°

0.7 0.6 1.0

r

0.5

0.0

0.5

r

1.0 1.0

r

0.5

0.0

0.5

r

1.0

1.2

u 1.0 0.9 0.8

T = 270°

T = 135°

T = 90°

T = 315°

0.7 0.6 1.0

r

0.5

0.0

0.5

r

1.0 1.0

r

0.5

0.0

0.5

r

1.0

0.1

w

T = 180°

T = 0°

T = 225°

T = 45°

0.0

-0.1 1.0

r

0.5

0.0

0.5

r

1.0 1.0

r

0.5

0.0

0.5

r

1.0

0.1

w

T = 270°

T = 90°

T = 135°

T = 315°

0.0

-0.1 1.0

r

0.5

0.0

0.5

r

1.0 1.0

r

0.5

0.0

0.5

r

1.0

Fig. 3.2. Velocity distributions u(r, θ) and w(r, θ) 10 diameters downstream of a bend at Re = 1.9 × 105 . (◦) Measurements after Mickan (1999); (—–) Series expansions according to (3.1) and (3.2), k = 0, 1, 2; i = 1, 2, 3 for x = 0. The coefﬁcients are listed in Table 3.3. (- - - -) Series expansions according to (3.1) and (3.2). Minimal program: k = 0, 1, 2; i = 1 for x = 0.

3 Optimal Characteristic Parameters for the Disturbances in Turbulent Pipe Flow

W wx /(N/m2)

3

0.8

2

0.4

1

0

0 0°

90°

180°

270°

T

W w T /(N/m2)

58

-0.4 360°

Fig. 3.3. Circumferential distributions of the wall shear stress components after Wildemann (2000). Position: 11 diameters downstream of a single bend. Re = 2.2 × 105 ( = 0.0439). τw∞ = 2.28N/m2 (measured); () τwx measured; (◦) τwθ measured; (—–) Fourier series (terms up to k = 2).

the characteristic parameters. The measurements have been carried out 11 diameters downstream of a single bend. Since the geometries and the Reynolds numbers of the experimental investigations are very similar, the characteristic parameters of Mickan (1999) and Wildemann (2000) are compared with each other in Table 3.4. Only the intensity of the three-dimensional ﬂow structures characterized by ' 2 (3.30) C... = A2... + B... is considered. The agreement is surprisingly good except for CW11 . The value CW11 measured by Wildemann is probably the correct one because it agrees with other results found in the literature. Norman et al. (1989), for instance, measured the value Table 3.4. The optimal characteristic parameters describing the intensities of the ﬁrst six ﬂow structures given in Table 3.1 for the turbulent pipe ﬂow about 10 diameters downstream of bends at Re ≈ 2 × 105 . Comparison of data by Mickan (1999) and Wildemann (2000). Mickan (1999) Wildemann (2000) Re = 1.9 × 105 Re = 2.2 × 105 x/D = 10 x/D = 11 AU01 CU11 CU21 BW01 CW11 CW21

0.054 0.032 0.013 0.002 0.015 0.021

0.053 0.032 0.012 0.013 0.070 0.021

References

59

wmax (= 0.12) at the axis 6.2 diameters downstream of a bend at Re = 10 5 . This leads according to (3.23), (3.24) and (3.30) to CW11 =

0.12 |wmax (0)| = = 0.11 |WW11 | 1.1

(3.31)

which is close to Wildemann’s result.

References Cebeci T, Chang KC (1978) A general method for calculating momentum and heat transfer in laminar and turbulent duct ﬂows. Numerical Heat Transfer 1:39–68 Gersten K, Merzkirch W, Wildemann C (2001) Verfahren und Vorrichtung zur Korrektur fehlerhafter Messwerte von Durchﬂussmessgeräten infolge gestörter Zuströmung. Patent No. 197 24 116 Klein A (1981) REVIEW: Turbulent developing pipe ﬂow. J. Fluids Eng. 103:243–249 Mickan B (1999): Systematische Analyse von Installationseffekten sowie der Efﬁzienz von Strömungsgleichrichtern in der Großgasmengenmessung. Doctor thesis, University Essen Morrison GL, DeOtte RE, Beam EI (1992) Installation effects upon oriﬁce ﬂow meters. Flow Meas. Instrum. 3:89–93 Morrison GL, Hauglie I, DeOtte RE (1995) Beta ratio, axisymmetric ﬂow distortion and swirl effects upon oriﬁce ﬂow meters Flow Meas. Instrum. 6:207–216 Norman RS, Mattingly GE, McFaddin SE (1989) The decay of swirl in pipes. Proceedings International Gas Research Conference 1989:312–321 Steenbergen W, Voskamp J (1998) The rate of decay of swirl in turbulent pipe ﬂow. Flow Meas. Instrum. 9:67–78 Sudo K, Sumida M, Hibara H (1998) Experimental investigation of turbulent ﬂow in a circularsectioned 90-degree bend. Exp. in Fluids 25:42–49 Wildemann C (2000) Ein System zur automatischen Korrektur der Messabweichung von Durchﬂussmessgeräten bei gestörter Anströmung. Doctor thesis, Universität Essen. See also: Fortschritt-Berichte VDI, Reihe 8, Nr. 868 Zagarola MV, Smits AJ (1998) Mean-ﬂow scaling of turbulent pipe ﬂow. J. Fluid Mech. 373:33–79

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners Wei Xiong, Kathrin Kalkühler, Wolfgang Merzkirch Institut für Strömungslehre, Universität Essen, 45117 Essen, Germany

Particle-image velocimetry and hot-wire anemometry are used for experimentally studying the ﬂow downstream of three conditioners, a tube bundle and two perforated plates at Reynolds numbers of the order 105 . The conditioners are exposed to the ﬂow disturbed by two different installations: a 90◦ single bend and a 2 × 90◦ out-of-plane double bend. Velocity proﬁles, turbulent ﬂuctuations and Reynolds’ stress are measured. The jets issuing from the holes and tubes of the conditioners are visualised in the near ﬁeld, which extends to approximately four pipe diameters downstream of the conditioners. At that position the disturbance imposed on the ﬂow by the conditioners disappears, while the decay of the disturbance caused by the installations takes place in the far ﬁeld. The decay rate in the far ﬁeld depends on the speciﬁc installation. While the velocity proﬁles match the proﬁle for fully developed ﬂow at a position of approximately 25 diameters downstream of the conditioners, the turbulent equilibrium state is not reached even at 50 diameters. The results also show that the perforated plates have a higher efﬁciency than the tube bundle in conditioning the disturbed ﬂow.

4.1 Introduction Flow conditioners serve for homogenising a velocity proﬁle in pipe ﬂow that is disturbed by an installation. A certain class of conditioners is intended to accelerate the formation of a fully developed turbulent velocity proﬁle, i.e., a deﬁnite state of ﬂow that must exist at the position of a ﬂow meter whose calibration was done in fully developed ﬂow. Such conditioners, a number of which are deﬁned in national or international standards, e.g., DIN1952 (1982), allow to reduce the length of straight pipe required upstream of a ﬂow meter. This obvious advantage must be paid for by an additional pressure decrease, expressed, e.g., by a pressure “loss” coefﬁcient ξ = (2 ∆p)/(ρ u2b ), with ∆p being the pressure decrease caused by the conditioner, ρ the ﬂuid density, and ub the bulk velocity in the pipe (volume ﬂow divided by the cross-sectional area). Many attempts have been made in optimising the design of a conditioner such that the desired reduction in straight pipe length is maximized

62

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

and the inevitable pressure decrease ∆p minimised. This led to a variety of different approaches in the design of ﬂow conditioners that are currently in practical use. Here we report on measurements of the distribution of ﬂow velocity and turbulence downstream of three different conditioners: tube bundle (see (DIN1952, 1982)), and two perforated plates according to the design of Akashi et al. (1978) and Laws (1990). The applied experimental methods are hot-wire and particle-image velocimetry (PIV). The PIV measurements are an extension of experiments reported earlier by Schlüter and Merzkirch (1996) with considerably improved experimental capabilities: The tracer particles being much smaller than those used in the earlier experiments allow to resolve velocity ﬂuctuations; time-averaged velocity proﬁles can be determined from a total of several hundred digitally recorded PIV images; PIV recordings can be taken in measurement planes normal to the pipe axis, as speciﬁed in detail in Xiong and Merzkirch (1999). The independent use of two different experimental methods and the comparison of the data acquired with them improves the conﬁdence in the measurement results. The aim of the measurements is to characterize the ﬂow ﬁeld downstream of the conditioners that are exposed to pipe ﬂow disturbed by two speciﬁc installations: a 90◦ bend and a 2 × 90◦ out-of-plane double bend. It is of interest to see how quickly the ﬂow approaches the fully developed state, regarding time-averaged velocity proﬁles and turbulence characteristics. This equilibrium state is formed in the far ﬁeld of the conditioners. The speed of its formation should depend on the rate at which the ﬂuid jets issue from the tubes or openings mix in the near ﬁeld of the conditioners. Therefore, the near ﬁeld is also investigated and visualised with PIV. The Reynolds number ReD in the experiments is of the order of 100.000, based on the pipe diameter D. This magnitude of ReD is of interest for many technical applications, and it permits to compare the results with other data reported in the literature. A great number of these results were obtained for the combination of a ﬂow conditioner with a speciﬁc ﬂow meter. The investigations reported by Morrison et al. (1997) are a typical example of respective research for this order of magnitude of the Reynolds number. Extensive descriptions of many details of the experimental programme reported in this chapter are given in the dissertations of Kalkühler (1998) and Xiong (2000).

4.2 Experiments The test rig is a steel pipeline with D = 100 mm inner diameter and hydraulically smooth internal surfaces. Several straight pipe sections can be assembled to provide a total length of straight pipeline of 120 D, if so desired. Air as the working ﬂuid is sucked through the pipe by means of a ventilator positioned at the downstream end of the line. The highest achievable Reynolds number, approx. 300,000, is determined by the power of the ventilator. The installations are positioned 23 D downstream of the intake nozzle, while the ﬂow conditioners are at a distance of 2 D downstream of the installations (Fig. 4.1). The distance of straight pipe between installation and

4.2 Experiments

63

ventilator can be up to 80 D. The test rig is equipped with a device for taking continuous reference measurements of the volumetric ﬂow rate with an uncertainty of ±0.3%: further details of the test rig and the reference measurements can be found in Kalkühler (1998) and Xiong (2000).

Fig. 4.1. Pipe line test rig.

Like in many comparable experimental investigations two installations are used in the experiments: A 90◦ bend with D = 100 mm i.d. and bend radius RB = 2 D, and a 2 × 90◦ out-of-plane double bend, composed of two 90◦ single bends. The three ﬂow conditioners are depicted in Fig. 4.2.

Fig. 4.2. The three conditioners investigated: (a) tube bundle , (b) perforated plate according to Akashi et al. (1978), (c) perforated plate according to Laws (1990).

An optical test section for taking PIV measurements can be installed at selected distances downstream of the conditioners. Two types of PIV test sections are used that allow different spatial orientation of the light sheet plane. For measuring the axial and radial components of the ﬂow velocity in the pipe, the light sheet is vertical and includes the (horizontal) pipe axis. The motion of the tracer particles is observed and recorded by a CCD camera that is directed normal to the plane of the light sheet,

64

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

i.e., in horizontal direction. Optical access for the observation is provided through the test section’s transparent wall made of 0.3 mm thick acetate sheet so that optical distortions are minimised (Fig. 4.3a).

Fig. 4.3. Test sections used for the PIV experiments: (a) axial and radial velocity components are measured in the vertical light sheet that is observed through the acetate sheet forming part of the pipe wall; (b) radial and azimuthal velocity components are measured in the light sheet arranged normal to the pipe axis; observation is through the endoscope in upstream direction.

The radial and azimuthal velocity components (e.g., for determining swirl) are measured with an optical test section that allows to position the light sheet normal to the pipe axis (Fig. 4.3b). PIV recordings are taken through an endoscope located at 1.5 D downstream of the light sheet, with the viewing direction upstream. With a light sheet thickness of 1.2 mm it is secured that a sufﬁcient number of tracer particles is illuminated twice by the laser double pulse while moving within the light sheet. Since the particles are imaged under different viewing angles, an optical correction is necessary for determinig the true values of the radial and azimuthal velocity components (Xiong and Merzkirch, 1999). Oil droplets as tracer particles are added to the air at the inlet nozzle of the test rig. Phase Doppler measurements show that these oil droplets have a size distribution with a maximum between 3 and 4 µm (Xiong, 2000). Illumination for the PIV measurements is provided by a double-pulsed Nd:YAG laser. The two successive particle-images are recorded with a CCD camera on separate frames, thus allowing an evaluation by means of the cross-correlation technique. The synchronisation of the laser’s pulse repetition rate and the camera’s frame rate results in a total of 6 PIV recordings that can be taken per second, practically at any desired length of measure-

4.3 Results and Discussion

65

ment period. This allows to determine time-averaged velocity values from a total of several hundred individual and statistically independent PIV recordings. A DANTEC constant-temperature hot-wire anemometer with a crossed-wire probe is used for velocity and turbulence measurements. Calibration and measurments are controlled by a PC and the software “Streamline”. In the calibration unit the probe is exposed to 13 different values of the ﬂow angle between −30 ◦ and +30◦ , and to 15 different values of the ﬂow velocity ranging from 0.5 m/s to 55 m/s. The results of the calibration are conserved in a look-up table which is also used, with respective interpolations, for the evaluation of the measurements. The hot-wire measurements are performed using a special test section that can be inserted into the pipeline at selected positions along the pipe axis. The probe is traversed along radial paths separated by constant angular intervals, and measurements are taken at steps of 2 mm along each path and with the probe at three different orientations (0◦ , 45◦ , 90◦ ) relative to the probe axis. This way, the three Carthesian velocity components can be determined and then converted into the axial, radial and azimuthal components. Two orientations of the probe (0 ◦ , 90◦ ) are, in principle, sufﬁcient for this purpose; the third orientation at 45 ◦ serves for increasing the measurement accuracy. The anemometer system and the applied software allow to determine mean (time-averaged) values of the velocity, RMS values of the turbulent ﬂuctuations, and Reynolds stresses; further details are reported by Kalkühler (1998).

4.3 Results and Discussion 4.3.1 Nearﬁeld Downstream of the Conditioners The conditioner is a partial obstruction in the pipe cross section, therefore causing the mentioned pressure decrease, and the ﬂuid is released to the downstream side through the openings or tubes from which the ﬂuid issues in form of jets. Desirable is an intensive and quick mixing of these jets. We consider the near ﬁeld to extend in downstream direction to an axial position where the jets have mixed completely and individual jets can no longer be discriminated. The measurement results presented in the following show that, for the range of Reynolds numbers investigated, this axial extension of the near ﬁeld is approximately 4 D. Proﬁles of the time-averaged axial velocity, measured with PIV in the near ﬁeld of the three conditioners, are shown in Fig. 4.4. The vertical light sheet includes the pipe axis, and its orientation with respect to the exit plane of the conditioners is indicated on top of this ﬁgure. The Reynolds number is Re D = 100,000. The ﬂow is disturbed, upstream of the conditioner, by the 2 × 90 ◦ out-of-plane double bend. The velocity proﬁles exhibit two patterns characteristic for this conﬁguration: the peaks indicating the position of the jets whose origin is obvious for the given orientation of the light sheet, and an asymmetry (skewness) caused by the double bend. For the purpose of comparison and reference, the diagrams also include the velocity proﬁles that would exist at this position in the case without conditioner, and the fully developed proﬁle for this Reynolds number, calculated with the theory of

66

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

Fig. 4.4. Time-averaged proﬁles of the non-dimensional axial velocity component u in the near ﬁeld of the three conditioners that are exposed to the ﬂow disturbed by the 2 × 90◦ out-ofplane double bend; ub = bulk velocity, r = radial coordinate, D = pipe diameter. Orientation of the light sheet with respect to the conditioner plane is indicated at the top. Fully developed proﬁle (solid line) and the proﬁle that would exist at the measurement position in the absence of a conditioner (open circles) are shown as reference. ReD = 100,000.

Gersten and Herwig (1992). The jets can be considered as a disturbance imposed to the ﬂow by the conditioner, and it is visible that this disturbance decays rapidly, in contrast to the disturbance caused by the installation, the skewness, that persists clearly at the downstream end of the near ﬁeld where the jets are no longer detectable. Thus, the self-generated disturbance of the conditioner disappears at the end of the near ﬁeld, while the disturbance caused by the installation must decay in the far ﬁeld, i.e., for x/D > 4. At the end of the near ﬁeld there is no signiﬁcant difference in the form of the velocity proﬁles, i.e., in the performance of the three conditioners investigated.

4.3 Results and Discussion

67

Fig. 4.5. Instantaneous distribution of the axial and radial velocity components in the near ﬁeld of the three conditioners: (a) tube bundle, (b) perforated plate according to Akashi et al. (1978), (c) perforated plate according to Laws (1990). Orientation of the light sheet and installation upstream of the conditioner as in Fig. 4.4; ReD = 100,000. Velocities are depicted by arrows and velocity ranges by shades of gray.

Instantaneous velocity distributions measured immediately downstream of the conditioners, in the axial range from 0.2 ≤ x/D ≤ 0.7, are presented in Fig. 4.5. The orientation of the PIV light sheet is indicated in Fig. 4.4; the left side of the velocity plots in Fig. 4.5 corresponds to the upper half of the light sheet plane as traced in Fig. 4.4. Form and position of the visible jets can be related easily to the positions of the openings of the conditioners from which these jets issue. Figure 4.5 provides an impression on the mixing of the jets in the regime covered by this PIV recording. This mixing is governed by interactions of the initially parallel jets. Until now, such interaction processes have been investigated only for two or three

68

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

parallel jets (e.g., Nasr and Lai 1997). From these investigations it is known that the jet interaction causes an unsteadiness of the whole regime, and it must be expected that the jets in the near ﬁeld of the conditioners tend to tumble around their axes. The applied PIV technique, however, does not provide the temporal resolution necessary for verifying this assumption.

Fig. 4.6. Instantaneous pattern of the turbulence structure for the case shown in Fig. 4.5. A constant axial velocity of 15 m/s is subtracted from the distributions in Fig. 4.5.

Visualisation of the instantaneous distribution of the turbulent ﬂuctuations in the measurement plane would require to ﬁrst determine time-averaged values of the ve-

4.3 Results and Discussion

69

locity, and then to subtract this distribution from an instantaneous velocity record, like the one in Fig. 4.5. Because of the mentioned unsteadiness of the whole ﬂow, such time-averaging does not make sense, since both turbulence and unsteadiness would be averaged together. Instead, the constant value of the bulk velocity, 15 m/s (in axial direction), is subtracted from the velocity patterns of Fig. 4.5, in order to obtain a rough impression of the distribution of the turbulent ﬂuctuations. The result is presented in Fig. 4.6 which gives an indication of the strong ﬂuid motion normal to the jet axes. This motion which is a measure of the lateral mixing is veiled in the velocity plots of Fig. 4.5 by the dominating axial velocity component. A comparison of the rates of mixing caused by the three different conditioners appears not possible due to the different association of the light sheet plane to the distribution of openings in the exit plane of the conditioners. 4.3.2 Redevelopment of the Flow in the Far Field The further decay of the ﬂow disturbances in the far ﬁeld of the conditioners is measured with hot-wire anemometry and PIV. Proﬁles of the time-averaged axial velocity component, measured in the vertical plane through the pipe axis (cf. Fig. 4.4), with the hot-wire anemometer are presented in Fig. 4.7 for the axial range 5.5 ≤ x/D ≤ 51.5 downstream of the conditioners. As in the case of the previous PIV examples the installation upstream of the conditioners is the 2 × 90 ◦ out-ofplane double bend. The proﬁles for x/D = 5.5 and 11.5 are measured with both PIV and hot-wire, and the patterns are practically identical. The skewness in the proﬁles, as it is visible at the end of the near ﬁeld, disappears at approximately x/D = 20 downstream of the two perforated plates, while a residual skewness is still visible downstream of the tube bundle. A comparison of the performance of the investigated conditioners, based on the measurement of the axial velocity component as results shown in Fig. 4.7, would require to compare the proﬁles for not only one orientation of the measurement plane. Four planes are used in the measurements, each including the pipe axis: 0◦ (horizontal), 45◦ , 90◦ (vertical), 135◦. For the purpose of comparing the conditioners we determine mean values of the deviation (in percent) of the measured velocity from the value for fully developed ﬂow along each measurement path (diameter according to the orientation of the measurement plane). The result applying to the 2 × 90◦ out-of-plane double bend as the installation upstream of the conditioners is shown in Fig. 4.8. At x/D = 11.5 the relative deviation is approximately 3% for the two perforated plates (Laws and Akashi), while it is between 3% and 5%, depending on the measurement plane, for the tube bundle. The dependence of the deviation on the orientation of the measurement plane as it is obvious for the tube bundle again shows, that the skewness of the velocity proﬁle has not yet disappeared. The values given here in % do not indicate the possible deviation of a ﬂow meter’s reading from the true value of the volumetric ﬂow rate if the meter would be placed at that axial position: it is not known how a speciﬁc meter reacts on the disturbance of the velocity proﬁle.

70

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

Fig. 4.7. Proﬁles of the time-averaged, non-dimensional axial velocity component u/ub in the far ﬁeld of the three conditioners that are exposed to the ﬂow disturbed by the 2 × 90◦ out-of-plane double bend (hot-wire data); ub = bulk velocity in the pipe (volumetric ﬂow rate/area of pipe cross section), r = radial coordinate, D = pipe diameter. Orientation of the light sheet with respect to the conditioner plane is as indicated in Fig. 4.4. Fully developed proﬁle is shown for comparison (solid line); ReD = 100,000. (a) perforated plate according to Laws (1990), (b) perforated plate according to Akashi et al. (1978), (c) tube bundle.

Figure 4.8 also indicates the relative deviation of the axial velocity proﬁle from the fully developed state for the case that no conditioner is used. At x/D = 11.5 this value is approximately 6%. The hot-wire measurements show that, downstream of the three conditioners, the mean deviation is smaller than 2% for x/D ≥ 25. It is believed that the accuracy in determining this value is 1% to 2%, so that the given value indicates practically the redevelopment of the undisturbed state, at least regarding the velocity proﬁles. It is known that the 2 × 90◦ out-of-plane double bend produces swirl. As stated above, the swirl components of the velocity ﬁeld can be measured by means of the PIV setup with the light sheet perpendicular to the pipe axis (Xiong and Merzkirch, 1999), and also by hot-wire anemometry. As an example the instantaneous distribution of the radial and azimuthal velocity components at a position x/D = 1.5 downstream of the bend and for the Reynolds number Re D = 100,000 is shown in Fig. 4.9. Besides the over-all swirl one recognises the two local vortices of the “sec-

4.3 Results and Discussion

71

Fig. 4.8. Relative deviation in % (deﬁnition see text) from fully developed ﬂow downstream of conditioners that are exposed to the ﬂow disturbed by the 2 × 90◦ out-of-plane double bend: (a) tube bundle, (b) perforated plate according to Akashi et al. (1978), (c) perforated plate according to Laws (1990); (d) values for the case without conditioner. Results for the measured axial velocity are shown for four orientations of the measurement plane, each including the pipe axis: 0◦ (horizontal), 45◦ , 90◦ (vertical), 135◦ . ReD = 100,000.

ondary” ﬂow generated in the bend. The positions of these secondary vortices are marked by circles in the ﬁgure. The degree at which a velocity distribution in the pipe is disturbed by swirl can be expressed by the “swirl angle”, as deﬁned by Mattingly and Yeh (1989). From both hot-wire and PIV measurements we determine the “mean” swirl angle, i.e., the swirl angle averaged over the respective cross section in the pipe, in the ﬂow downstream of the three conditioners. As an example the result for the conditioner of Laws (1990) is shown in Fig. 4.10 together with the values of the mean swirl angle at the same axial positions as it is measured in the absence of any conditioner. The swirl is practically zero, even for x/D = 1.5, and this applies for all three conditioners investigated. In Fig. 4.10 the values of the swirl angle measured with conditioner are around 1◦ , and this is deﬁnitely below the value 2◦ that the norms, e.g., DIN1952 (1982), tolerate for measurements of the volumetric ﬂow

72

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

Fig. 4.9. Instantaneous distribution of the radial and azimuthal velocity components at 1.5 D downstream of the 2 × 90◦ out-of-plane double bend (PIV data in the y-z plane, normal to pipe axis). Positions of two vortices caused by the “secondary ﬂow” are indicated by circles.

rate. As a result regarding swirl one can state that the three conditioners remove swirl completely. Velocity ﬂuctuations are determined with the hot-wire anemometer in the axial range 5.5 ≤ x/D ≤ 51.5 downstream of the conditioners. RMS values of the axial component of the ﬂuctuation, measured in the vertical plane and nondimensionalised with the shear velocity ut, are presented in Fig. 4.11, again with the 2 × 90◦ out-of-plane double bend as installation and for ReD = 100,000. The distribution for fully developed ﬂow according to Lawn (1971) is given for comparison. Obviously the proﬁles for the tube bundle approach this equilibrium value more

Fig. 4.10. Swirl angle as deﬁned in Mattingly and Yeh (1989) downstream of position of conditioner; (a) with Laws (1990) conditioner, (b) without conditioner, (c) swirl angle 2◦ tolerated by norms. ReD = 100,000.

4.3 Results and Discussion

73

Fig. 4.11. RMS values of non-dimensional axial velocity ﬂuctuations, u /uτ , downstream of the three conditioners exposed to the ﬂow disturbed by the 2 × 90◦ out-of-plane double bend (hot-wire data): (a) perforated plate according to Laws (1990), (b) perforated plate according to Akashi et al. (1978), (c) tube bundle. Vertical measurement plane, ReD = 100,000. The proﬁle for fully developed ﬂow is shown as reference (solid line).

slowly than the proﬁles for the two perforated plates. Even at x/D = 21.5 there is a noticeable deviation from the equilibrium proﬁle, and agreement with fully developed ﬂow is only given, within the range of measurement accuracy, for x/D = 51.5. Similar results are found from the distribution of the Reynolds’ stress formed with the axial and radial ﬂuctuation (Fig. 4.12). Figures 4.11 and 4.12 indicate that the state of turbulence redevelops less rapidly than the velocity proﬁles, which can be explained with the additional production of turbulence by the conditioners. These results should be of interest for the modeling of the turbulence ﬁeld in respective numerical computations, and also for the possible use of ﬂow meters for which the generation of the signals depends on the turbulent state of the pipe ﬂow, e.g., meters based on ultrasound. Experiments are also performed with the 90◦ single bend as the installation upstream of the conditioners. Here, only two results typical for this series of measurements are presented. Further data can be found in Kalkühler (1998) and Xiong (2000). Mean values of the deviation of the time-averaged axial velocities from the

74

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

ucucW ur

2

ucucW ur

2

ucucW ur

2

ucucW ur

2

Fig. 4.12. Non-dimensional Reynolds’ stress formed with the axial and radial ﬂuctuation, u and ur , downstream of the three conditioners. Flow, notation and reference as in Fig. 4.11.

fully developed proﬁles, deﬁned above in connection with Fig. 4.8, are presented in Fig. 4.13. At x/D = 11.5 these deviations are somewhat higher than for the case with the 2 × 90◦ out-of-plane double bend, between 3% and 4% for the two perforated plates, and around 6% for the tube bundle. The latter value is even not better than for the case without conditioner. In general, the disturbance in the velocity ﬁeld caused by the single bend is damped to a lower degree by the conditioners than the disturbance produced by the double bend. Of course, this result applies here to the range of Reynolds numbers investigated. As a second example for the single bend we show in Fig. 4.14 the proﬁles of the Reynolds’ stress, equivalent to the case reported in Fig. 4.12. These data are determined with the hot-wire anemometer. They conﬁrm the conclusion that the three conditioners are less efﬁcient in removing the disturbance produced by the single bend in comparison to the case with the double bend. Even at x/D = 51.5, where the velocity proﬁles match well with the proﬁle for fully developed ﬂow (not shown here), there is a considerable difference to the Reynolds’ stress proﬁle for the equilibrium state. Reynolds numbers higher than 300,000 cannot be produced due to the limitation in the power of the ventilator used in the experimentel setup. The experiments performed with ReD = 300,000 show no signiﬁcant differences to the results already

4.4 Conclusion

75

Fig. 4.13. Relative deviation (deﬁnition see text) from fully developed ﬂow downstream of conditioners that are exposed to the ﬂow disturbed by the 90◦ single bend. Notation and reference as in Fig. 4.8.

presented; for details the reader is again referred to Kalkühler (1998) and Xiong (2000).

4.4 Conclusion The experimental results indicate, how rapid the ﬂow that is disturbed by two speciﬁc installations redevelops under the inﬂuence of the three conditioners investigated, the perforated plates according to Akashi et al. (1978) and Laws (1990), respectively, and the tube bundle as deﬁned in technical norms, e.g., DIN1952 (1982). The experiments are performed at Reynolds numbers of the order 10 5 . They show that the ﬂow downstream of the conditioners can be separated in two regimes: The disturbance imposed on the ﬂow by the conditioner disappears at the end of the near ﬁeld that extends to approximately four pipe diameters downstream of the conditioner. In this regard, the conditioners, particularly the perforated plates, can be compared with a turbulence grid. Such grids are often used for generating homogeneous and quasi-

76

4 Measurement of Velocity and Turbulence Downstream of Flow Conditioners

Fig. 4.14. Non-dimensional Reynolds’ stress formed with the axial and radial ﬂuctuation, u and ur , downstream of the three conditioners with the 90◦ single bend as the upstream installation. Notation and reference as in Fig. 4.10.

isotropic turbulence which is assumed to establish about 30 mesh sizes downstream of the grid (see, e.g., Hinze (1975)). From the geometry of the perforated plates (Fig. 4.2) one can derive an equivalent mesh size of approximately 15 mm, and the end of the near ﬁeld then coincides with the establishment of quasi-isotropic turbulence, i.e., a ﬂow state in which any information on the characteristic length scales of the grid or plate has disappeared due to dissipation. The decay of the disturbance caused by the installation takes place in the far ﬁeld. The decay rate is different for the two installations: the decay is more rapid for the double bend. The velocity proﬁles match the proﬁle for fully developed ﬂow approximately at 25 diameters downstream of the conditioners, which is in agreement with results reported by Morrison et al. (1997). However, the turbulence ﬁeld, as expressed e.g., by the Reynolds stress, does not reach the equilibrium state even at 50 diameters. This must be considered when describing this ﬂow numerically. A comparison of the three conditioners shows that the efﬁciency of the two perforated plates is higher than that of the tube bundle. The experiments do not yield direct information on how the reading of a ﬂow meter is affected by the deviation of the ﬂow from the fully developed state. If it is concluded that a meter must be positioned at 25 diameters or more downstream of the

References

77

conditioner, an alternative to a conditioner for avoiding very long lengths of straight pipeline might be the correction of the meter’s reading as proposed by Wildemann et al. (2002); see also Chap. 14 of this volume.

References Akashi K, Watanabe H, Koga K (1978) Flow rate measurement in pipe line with many bends. Mitsubishi Heavy Ind 15:87–96 DIN 1952 (1982) Durchﬂussmessung mit Blenden, Düsen und Venturirohren in voll durchströmten Rohren mit Kreisquerschnitt. Beuth-Verlag, Berlin Gersten K, Herwig H (1992) Strömungsmechanik - Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig Hinze JO (1975) Turbulence, 2nd edition. McGraw-Hill, New York Kalkühler K (1998) Experimente zur Entwicklung der Geschwindigkeitsproﬁle und Turbulenzgrößen hinter verschiedenen Gleichrichtern. Dissertation, Universität Essen; also published as VDI Fortschritt-Bericht, Reihe 7, Nr. 339, 1998, VDI-Verlag, Düsseldorf Lawn C (1971) The determination of the rate of dissipation in turbulent pipe ﬂow. J Fluid Mech 48:477–505 Laws EM (1990) Flow conditioning – a new development. Flow Meas Instrum 1:165–170 Mattingly G, Yeh T (1989) Flowmeter installation effects - single and double elbow installations. VDI Berichte 768:65–73 Morrison GL, Hall KR, Holste JC, Ihfe L, Gaharan C, DeOtte Jr RE (1997) Flow development downstream of a standard tube bundle and three different porous plate ﬂow conditioners. Flow Meas Instrum 8:61–76 Nasr A, Lai J (1997) Two parallel plane jets: mean ﬂow and effects of acoustic excitation. Exp. Fluids 22:251–260 Schlüter T, Merzkirch W (1996) PIV measurements of the time-averaged ﬂow velocity downstream of ﬂow conditioners in a pipeline. Flow Meas Instrum 7:173–179 Wildemann C, Merzkirch W, Gersten K (2002) A universal, nonintrusive method for correcting the reading of a ﬂow meter in pipe ﬂow disturbed by installation effects. J Fluids Eng 124:650–656 Xiong W, Merzkirch W (1999) PIV experiments using an endoscope for studying pipe ﬂow. J Flow Visual & Image Processing 6:167–175 Xiong W (2000) PIV-Untersuchungen im Nahfeld von Strömungsgleichrichtern. Dissertation, Universität Essen; also published by Shaker-Verlag, Aachen (ISBN 3-8265-7575-X)

5 Signal Processing of Complex Modulated Ultrasonic Signals Volker Hans Institut für Mess- und Regelungstechnik, Universität Essen, 45117 Essen, Germany

Ultrasound is a natural choice for the measurement of turbulent ﬂow. The ultrasonic signal is complex modulated by the structures in the ﬂuid. The modulation frequency is low resulting in narrow sidebands to the high carrier frequency. The modulated signal can be separated by digital undersampling. By means of hardware- or softwareapplied Hilbert-transform the complex modulated signal can be determined for crosscorrelation or vortex measurements. Phase angles can exceed the limited range of the arctan function caused by offset or strong vortices in the ﬂuid. The densitydistribution function of the phase angle admits the reconstruction of the phase angle. Extended Kalman ﬁlters have been proved to be suitable for phase demodulation. Also, analog signal processing will be explained for an electronic Hilbert-transform and amplitude demodulation in vortex measurement. Measurement results are represented for cross-correlation and vortex measurements at low and high ﬂow velocities.

5.1 Introduction On examining the propagation of small pressure ﬂuctuations in a ﬂuid a high similarity between ﬂow turbulences and ultrasound can be determined (Lighthill, 1972; Poppen, 1997). This effect makes ultrasound a natural choice for the measurement and analysis of turbulent structures. An ultrasonic beam crossing a ﬂuid perpendicularly will be complex modulated by the turbulent structures in the ﬂuid. The modulation is connected with the size of structures and the wavelength of the ultrasound. Usually the carrier frequency of the ultrasonic wave is in the range of some hundred kHz whereas the bandwidth of the modulation frequency comes up to some kHz. This means that the sidebands of the modulated signal are very close to the carrier frequency. Therefore, signal processing is of great importance for ultrasound measurements. Ultrasonic measurements in a gas ﬂow aim at the determination of the mean ﬂow velocity. Two principles are described in Chaps. 6 and 7 in detail. Vortex measurements are based on the well-known effect of the Kármán vortex street. The frequency

80

5 Signal Processing of Complex Modulated Ultrasonic Signals

of vortices in the wake of a bluff body is directly proportional to the mean ﬂow velocity. For signal processing the modulation frequency is to be determined. Correlation measurements as described in Chap. 7 require the determination of cross-correlation functions of two amplitude- or phase-modulated ultrasonic signals. In both cases the information needed is in the sidebands of the complex modulated signal. For signal processing the demodulation of the signal is necessary as may be shown by the example of cross-correlation measurements.

5.2 Cross-Correlation Measurements The principle of cross-correlation ﬂow measurement is based on two parallel ultrasonic barriers sending continuous beams radially through the pipe with a ﬂowing ﬂuid as shown in Fig. 5.1. Transmitters

Flow direction

v

Receivers Fig. 5.1. Measurement setup.

The distance of the barriers is in the range of the pipe diameter. Inhomogeneous structures in the ﬂuid, such as turbulences, lead to a modulation pattern in the ultrasonic signal like a ﬁngerprint. These patterns are similar in both signals but time shifted according to the traveling time of the modulating structures between the barriers. The similarity of the two patterns is compared by the crosscorrelation function 1 Φ(τ ) = lim T →∞ 2T

+T s1 (t)s2 (t + τ )dt ,

(5.1)

−T

leading to a characteristic maximum of highest similarity at the traveling time. s 1 (t) and s2 (t) are the signals of the two barriers, τ is a time shift. But as the modulation depth is relatively low the self-similarity of the sinusoidal carrier signal is predominating the cross-correlation function in a higher peak. There is a risk of the modulation signal being suppressed. Figure 5.2 shows the cross-correlation function of a modulated signal (left) with carrier signal and without carrier signal (right).

5.3 Demodulation by Digital Undersampling

81

Cross-correlation function

Stochastic signal without superposed oscillation Amplitude

Amplitude

Stochastic signal with superposed oscillation

Time

Time

Fig. 5.2. Cross-correlation functions of stochastic signals with (left) and without (right) superposed oscillation such as a carrier or pulsation.

The ﬁgures demonstrate clearly the necessity of signal demodulation. In the following, various kinds of signal-processing methods are summarized that are described in detail in (Filips, 2003; Niemann, 2002; Poppen, 1997; Skwarek, 2000; Windorfer, 2001).

5.3 Demodulation by Digital Undersampling The demodulation is based upon a synchronous sampling at zero-point crossings with integer submultiples of the carrier frequency taking advantage of the repetition of the spectrum of a sampled signal. This can be illustrated by the Fourier-transformation of a sampled signal from the time to the frequency domain. In the time domain sampling represents a multiplication with Dirac impulses in intervals T s of the sampling frequency +∞ s(t) = s(nTs )δ(T + nTs ) . (5.2) −∞

In the frequency domain it can be described by a repeating spectrum S(f ) =

+∞

s(f − nfs ) ,

(5.3)

−∞

with fs the sampling frequency between two spectra. The gap between the repeating bands is equivalent to the sampling frequency. A subdivision of this frequency by an integer number shifts one of these bands to f = 0 Hz, Fig. 5.3. Then the redundant spectra can be suppressed by high-pass ﬁltering and the signal is demodulated to its original band (Skwarek, 2000; Skwarek and Hans, 2000). Naturally, the Nyquist theorem must be taken into consideration with respect to the sidebands that are now shifted to a low-frequency range. From experience this range is less than 5 kHz. Suppose a 200-kHz ultrasound carrier frequency is undersampled by a factor 10, which means 20 kHz. This sampling frequency fullﬁlls the Nyquist theorem up to

82

5 Signal Processing of Complex Modulated Ultrasonic Signals frequency domain

time domain modulated signal undersampled signal

t

ts carrier-synchronized undersampling impulses

fcarrier

f modulation

0

fs

f original spectrum

repeating spectrum

Fig. 5.3. Undersampling in time to frequency domains.

sideband frequencies of 10 kHz. This kind of digital processing is very time saving, memory saving and cost saving. It can be readily applied for the determination of correlation functions. The demodulation of simultaneously amplitude- and phase-modulated signals is well-known from telecommunications engineering as quadrature-amplitude demodulation (QAD) or complex demodulation (Unbehauen, 1997). It can be demonstrated in the complex domain, Fig. 5.4. The modulated signal is represented by a

Fig. 5.4. Amplitude- and phase-modulated signal in complex domain.

phasor with changing amplitude and phase. Undersampling stops the rotation of the phasor such as is observed with a synchronized stroboscope and turns it into a standing pointer. Only local phase and amplitude vary according to modulation. Now, amplitude and phase can be obtained from the real and imaginary parts of the signal as 2 2 (5.4) a(t) = Re{s(t)} + Im{s(t)} , ϕ(t) = arctan

Im{s(t)} . Re{s(t)}

(5.5)

But in practical applications the imaginary part cannot be measured directly. Therefore, the Hilbert-transform has to be used.

5.4 Digital Hilbert-Transform

83

5.4 Digital Hilbert-Transform 5.4.1 Undersampled Hilbert-Transform Each causal signal as the modulated carrier signal is convertible into an analytical signal (Bachmann, 1992; Otnes and Enochson, 1978). The Hilbert-transform deﬁnes the connection between the real and imaginary parts for analytical signals. Turning the phasor at t1 by 90 degrees to t2 the projection of the phasor on the imaginary axis at t1 corresponds to the projection on the real axis at t2 , Fig. 5.5. Im

s(t2)

s(t1) 90° φ Re{s(t2)}= Im{s(t1)}

Re

Re{s(t1)}

Fig. 5.5. Reconstruction of imaginary part by Hilbert-transform with 90◦ phase shifted samples.

For digital sampling (5.4) and (5.5) can be written as a(t) = Re{s(t1 )}2 + Re{s(t2 )}2 ,

ϕ(t) = arctan

Re{s(t2 )} . Re{s(t1 )}

(5.6)

(5.7)

In principle, this theoretical procedure leads to deviations because the ﬂuid moves, the structures are dissipating and the modulation changes in the time interval from t1 to t2 . But a numerical example will show that this effect is negligible. Suppose a carrier frequency of fc = 220 kHz as applied in practical experiments and the time period is 4.545 ms. A quarter period is 1.136 ms. During this time the ﬂuid moves only 5.7 µm at a mean ﬂow velocity of 20 m/s. It is not to be expected that this very small motion of the ﬂuid inﬂuences the measured result. Sampling the 220-kHz signal shifted by 90 degrees leads to a sampling frequency of 880 kHz resulting in expensive signal processing. It is obvious to apply the same undersampling procedure to the Hilbert-transform as to the carrier frequency. This undersampled Hilbert-transform must fulﬁll the Nyquist theorem with regard to the undersampled carrier frequency. In this time the ﬂow will move about 31 µm for the example above.

84

5 Signal Processing of Complex Modulated Ultrasonic Signals

5.4.2 Software-Based Hilbert-Transform The Hilbert-transform of a real signal s(t) is strictly deﬁned as sˆ(t) = H{s(t)} =

1 π

+∞

−∞

s(t) dt , t−τ

(5.8)

sˆ(t) = s(t) ∗ h(t) , with h(t) = and its Fourier-transform 1 πt

F

(5.9)

1 πt

(5.10)

! = −j sign(ω) ,

(5.11)

with the sign function for ω. The pure imaginary spectrum is constant for negative and positive frequencies. Thus, the Fourier-transform of the Hilbert-transform of the signal can be described as ˆ [s(t) ∗ h(t)] == −j sign [ωF (ω)] , Fˆ (ω) = H

(5.12)

which indicates that a Hilbert-transform can be determined when processing the signal shifted by 90 degrees.

xn

Multiplier

FFT length N

FFT-1 length N

Weight function Hn

Imag. part of Hn

1 Real part of Hn

0

N-1

n

-1 Weight function

Fig. 5.6. Diagram of Hilbert-transform using FFT.

yn

5.4 Digital Hilbert-Transform

85

The Hilbert-transform can be calculated by the convolution of the discrete Fourier-transformed signal with the weight function Hn , which can be described as ⎧ ⎨ 0 + j ∗ 0 for n = 0, N/2 Hn = 0 + j ∗ 1.0 for 1 ≤ n ≤ N/2 − 1 ⎩ 0 − j ∗ 1.0 for N/2 ≤ n ≤ N − 1 , as shown in Fig. 5.6 (Lin, 2003). The input xn of the system is the measured real part, while the output yn is the Hilbert-transform of xn . 5.4.3 Measurement Results Examples of measured and calculated imaginary parts of the demodulated signal are shown in Fig. 5.7 for ﬂow velocities of 2 m/s (left) and 10 m/s (right). 1.6

2

v=2m/s

1.4

v=10m/s 1.5

1.2 measured

measured

1

imaginary part

imaginary part

1 0.8 0.6 0.4

calculated

0.5

0

calculated

0.2 0

-0.5

-0.2 -0.4 0

2000

-1

4000 6000 8000 10000 12000 14000 16000 18000 sampling number

0

2000

4000 6000 8000 10000 12000 14000 16000 18000 sampling number

Fig. 5.7. Measured and calculated imaginary parts of the demodulated signal.

As explained in Chaps. 6 and 7 the effect of amplitude modulation is small in comparison to the phase modulation. Therefore, in further signal processing for determining the cross-correlation function usually the phase-demodulated signal is applied. Figure 5.8 shows a cross-correlation function of an amplitude-demodulated signal for two ﬂow velocities. 0.8

0.7

v=2m/s

cross-correlation coefficient

0.6 0.5

0.5

0.4

0.4 0.3 0.2

0

0.3 0.2

0.1

0.1 0

v=20m/s

0.6

cross-correlation coefficient

0.7

0.1

0.2

0.3

0.4 0.5 time (s)

0.6

0.7

0.8

0.9

0

0

10

20

time (ms)

30

40

50

Fig. 5.8. Cross-correlation function of amplitude-demodulated signal with digitally measured Hilbert-transform.

86

5 Signal Processing of Complex Modulated Ultrasonic Signals

The peaks indicating the traveling time of the modulating structures are low and sometimes not clear, involving the risk of misinterpretation. The results using the calculated Hilbert-transform for the same signals are shown in Fig. 5.9. 0.8

0.7

v=2m/s

cross-correlation coefficient

0.6 0.5

0.5

0.4

0.4 0.3 0.2

0.3 0.2

0.1

0.1 0

v=20m/s

0.6

cross-correlation coefficient

0.7

0

0.1

0.2

0.3

0.4 0.5 time (s)

0.6

0.7

0.8

0

0.9

0

10

20

time (ms)

30

50

40

Fig. 5.9. Cross-correlation function of amplitude-demodulated signal with calculated Hilberttransform.

The difference between digitally measured and calculated imaginary parts is obvious. For further comparison the cross-correlation function of the phasedemodulated signal is shown in Fig 5.10 . The results are remarkably similar. Only the peak of the cross-correlation function at higher velocity is smaller. 0.8

0.8

v=2m/s

cross-correlation coefficient

0.6 0.5 0.4 0.3 0.2 0.1 0

0

v=20m/s

0.7 cross-correlation coefficient

0.7

0.6 0.5 0.4 0.3 0.2

0.1

0.1

0.2

0.3

0.4 time (s)

0.5

0.6

0.7

0.8

0

0

10

20

30 time (ms)

40

50

Fig. 5.10. Cross-correlation function of phase-demodulated signal at different ﬂow velocities.

5.4.4 Analog Electronic Hilbert-Transform An alternative to digital-demodulation procedures are analog electronicdemodulation methods widespread in communication techniques (Mäusle, 1988). Bachmann (1992) and Otnes and Enochson (1978) suggest a principle as shown in Fig. 5.11.

5.4 Digital Hilbert-Transform

87

LP

LP

Fig. 5.11. Principle of an electronic Hilbert-transform.

Here, a modulated signal s(t) = a(t) sin(ωt + ϕ) is multiplied by the carrier and its 90◦ -shifted picture, respectively. They form the signals u(t) = a(t) sin (ωt + ϕ(t)) cos(ωt)

(5.13)

w(t) = a(t) sin (ωt + ϕ(t)) sin(ωt) ,

(5.14)

and which are transformed by trigonometric relations to 2u(t) = a(t) [sin ((ϕ(t)) + sin (2ωt + ϕ(t))]

(5.15)

2w(t) = a(t) [cos ((ϕ(t)) − cos (2ωt + ϕ(t))] .

(5.16)

and The low-frequency portion is separated with a low-pass ﬁlter so that u (t) = and

a(t) cos (ϕ(t)) 2

(5.17)

a(t) sin (ϕ(t)) (5.18) 2 remain without any inﬂuence of the carrier. Both parts, also called inphase and quadrature parts, respectively, represent a Hilbert-couple of a complex measure and can be further processed to (5.6) and (5.7). The already-mentioned problem becomes obvious again due to the calculation of tangent and squareroot functions with sufﬁcient accuracy. At this point the signals are sampled and computers are needed for ongoing calculations. An improved method for hardware-based demodulation is described in Skwarek and Hans (2001). w (t) =

88

5 Signal Processing of Complex Modulated Ultrasonic Signals

5.5 Phase Reconstruction On installing the ultrasonic transducers a correct adjustment is necessary, considering the phase offset. Small zero-phase shifts are unavoidable. Especially in vortex measurements, high phase angles can occur and in connection with the offset they can have angles greater than the range of the deﬁnition of the arctan function from −90 to +90 degrees. In this case, special algorithms for the reconstruction must be applied (Bachmann, 1992; Otnes and Enochson, 1978; Skwarek, 2000; Tribolet, 1977). A new kind of reconstruction has been developed by Niemann (2002) based on the analysis of the probability density distribution of the phase angle. The stochastic modulation of the ultrasonic beam by a streaming ﬂuid is normally distributed. But assuming no adjustment, the density function is shifted. The position of the maximum of the distribution function represents the zero-adjustment point of the receiving sensor. Figure 5.12 shows a demodulated phase signal with zero-adjustment error (left) and the corresponding distribution function (right). The zero-phase angle is about 150 degrees. time signal

distribution function

200 150

distribution

angle [°]

100 50 0 -50 -100 -150 -200

-150 -100 -50

Zeit

0

50 100 150

angle [°]

Fig. 5.12. Demodulated phase signal with zero-adjustment error (left), distribution function of the phase angle (right).

With the knowledge of the zero-adjustment point the phase signal can be relocated. The result of this technique is shown in Fig. 5.13. The time signal is normally distributed with a mean value of zero. time signal

distribution function

200 150

distribution

angle [°]

100 50 0 -50 -100 -150 -200

time

-150 -100 -50

0

50 100 150

angle [°]

Fig. 5.13. Demodulated phase signal after reconstruction (left), distribution function of the phase angle (right).

5.6 Phase Demodulation with Kalman Filter

89

The procedure results in the correct determination of the zero-adjustment point and the signal can be reconstructed correctly. The method is independent of the mechanical adjustment of the transducers and allows the measurement of ﬂow velocity even with disturbed proﬁles behind single and double elbows (Niemann and Hans, 2002).

5.6 Phase Demodulation with Kalman Filter The phase modulation of the ultrasonic wave by stochastically distributed structures in the streaming ﬂuid is a random process especially in the case of cross-correlation measurement. In vortex measurements the modulation is caused primarily by welldeﬁned vortices resulting in sinusoidal phase modulation superposed by random noise from other structures in the ﬂuid. Therefore it is obvious to process the modulated signal with recursive ﬁlters considering process noise and measurement noise. Kalman ﬁlters are well-known for this task describing linear dynamic systems in terms of state-space concepts. Nonlinearities can be taken into account by extended Kalman ﬁlters (EKF) (Haykin, 1996; Loffeld, 1990). In addition, the EKF has the great advantage that it can dispense with the zero-point adjustment of the sensors. An example for low modulation degree is given in Fig 5.14 (Filips, 2003). 140

120

120

Phase [°]

Phase [°]

140

100

100

80

0

0.1

0.2

Time [s]

0.3

0.4

80

0

0.1

0.2

Time [s]

0.3

0.4

Fig. 5.14. Phase reconstruction by QAD (left) and prediction EKF (right) at low modulation.

The phase reconstruction by EKF is compared with the QAD algorithm. There is very good similarity, conﬁrming the theory of EKF application. With increasing modulation the advantages of EKF become more obvious. Figure 5.15 shows an example. The phase reconstruction by QAD is superposed by high noise and must be processed by digital ﬁlters, whereas the result of extended Kalman ﬁlter is free of noise and can be directly processed.

5 Signal Processing of Complex Modulated Ultrasonic Signals 160

160

120

120

Phase [°]

Phase [°]

90

80

80

40

40

0.2

0.1

0

Time [s]

0.3

0.2

0.1

0

0.4

Time [s]

0.4

0.3

Fig. 5.15. Phase reconstruction by QAD (left) and prediction EKF (right) at high modulation.

To demonstrate the power of EKF, two complex modulated ultrasonic signals have been cross-correlated. The maximum of the cross-correlation function (CCF) represents the traveling time of structures between the two barriers. Figure 5.16 shows the cross-correlation function of a disturbed ﬂow at high ﬂow velocity of 20 m/s behind a single elbow. 1

0.8

Cross correlation correlation function function Cross

Cross correlation function

1

v = 20 m/s

Kalman

0.6

0.4

QAD

0.2

0.8

Kalman

v = 20 m/s

0.6

0.4

0.2 QAD

0

0

0.005

0.01

0.015

Time [s]

0.02

0.025

0.03

0

0

0.005

0.01

0.015

Time [s]

0.02

0.025

0.03

Fig. 5.16. Cross-correlation function of two demodulated ultrasonic phase signals by QAD and EKF without (left) and with (right) zero-point adjustment of the sensors.

In the left ﬁgure the QAD method shows a very low maximum of CCF in comparison to EKF. In this case, measurement was made without zero-point adjustment of the sensors. The right ﬁgure shows the same measurement with zero-point adjustment of the sensors. Comparing the two ﬁgures the advantage of the extended Kalman ﬁlter is evident. Similar results for undisturbed ﬂow as well as for disturbed ﬂow after double elbows could be obtained (Filips, 2003). Even vortex measurements could be performed successfully. The demodulated phase signal shows a periodical shape with noise in the time domain. But the phase angle often exceeds the range of the deﬁnition of the arctan function. Figure 5.17

5.7 Analog Signal Processing -70

-90

-90

Phase [°]

Phase [°]

-70

91

-110

-110

-130

0

0.1

0.2

Time [s]

0.3

-130

0.4

0

0.1

0.2

Time [s]

0.3

0.4

Fig. 5.17. Phase reconstruction by QAD (left) and prediction EKF (right) at low velocities.

200

300

100

100

Phase [°]

Phase [°]

shows the phase reconstruction for a M10 threaded control rod as a bluff body at 2 m/s ﬂow velocity. For low modulation grade the results of QAD and EKF algorithms are similar. Increasing ﬂow velocity leads to high modulation grades caused by increasing the rate of vortices and superposed by increasing noise, Fig. 5.18.

0

-300

-100

-200

-100

0

0.1

0.2

Time [s]

0.3

0.4

-500

0

0.1

0.2

Time [s]

0.3

0.4

Fig. 5.18. Phase reconstruction by QAD (left) and prediction EKF (right) at high velocities.

The resolution using QAD gives no explanation. Phase jumpings characterize the illustration. The reconstruction using an extended Kalman ﬁlter demonstrates a noise-reduced phase shape and yields better signal processing for the determination of the vortex frequency.

5.7 Analog Signal Processing Various analog demodulation techniques have been proved with more or less good results for disturbed or undisturbed ﬂow in correlation as well as in vortex measurements (Filips and Hans, 2002; Filips, 2003). A simple but very powerful method

92

5 Signal Processing of Complex Modulated Ultrasonic Signals Signal Signal generator generator ac Z c

transmitter s c (t ) LP

v

s d (t )

s AM (t ) receiver

Fig. 5.19. Analog signal processing in vortex measurement.

of amplitude demodulation especially for vortex measurements is pointed out in the following. The modulated receiver signal is multiplied by the feeding carrier signal, Fig. 5.19. For reasons of simpliﬁcation only an amplitude-modulated signal may be considered, which is expressed by sAM (t) = ac sin (ωc t) +

am am cos ((ωc − ωm ) t) − cos ((ωc + ωm ) t) , (5.19) 2 2

with ac , am amplitudes of carrier and modulating signal and ωc , ωm corresponding frequencies. This signal, multiplied by the carrier signal, is s(t) = sAM (t) ac sin (ωc t)

(5.20)

and results, according to trigonometric formulae, in

s(t) =

am ac ac 2 [1 − cos (2ωc t)] + sin (ωm t) 2 2 am ac [sin ((2ωc − ωm ) t) − sin ((2ωc + ωm ) t)] . + 4

v=5m/s

(5.21)

v=10m/s

Fig. 5.20. Receiver signal (lower signal) and signal after multiplication and ﬁltering (upper signal). Bluff-body width 4 mm, length 12 mm, tip to the inﬂow.

5.8 Conclusion

93

The signal contains the low-frequency modulating part with the modulating frequency ωm and high-frequency parts with the double carrier frequency that can be separated by a low-pass ﬁlter (LP). Figure 5.20 shows oscillograms from the modulated receiver signal (lower signal) and the demodulated signal after multiplication and ﬁltering that is caused directly by the vortices. The ﬁgure shows the signals for two velocities of 5 m/s and 10 m/s in the wake of a bluff body facing the tip to the inﬂow. The signals are nearly sinusoidal in shape. The vortex frequency as measured for the ﬂow velocity can be determined directly. The same procedure can be applied in the cross-correlation measurement, see Chap. 7. An example of the amplitude-modulated signal of one channel is shown in Fig. 5.21 for two velocities. On account of the weak signal the cross-correlation function is not so effective as that with phase-demodulated signals.

v=5m/

v=25m/s

Fig. 5.21. Signals of one channel for cross-correlation measurements. Lower signal: receiver signal; middle signal: product; upper signal: analog output.

5.8 Conclusion Processing of high-frequency complex modulated ultrasonic signals requires special procedures of demodulation. In the case of ﬂow-velocity measurement the ultrasonic signal is modulated by the structures in the ﬂuid. The bandwidth of modulation is up to 5 kHz. Digital undersampling methods have been proved to be suitable to separate the modulated signal from its carrier. The amplitude and phase angle of the complex signal can be measured by normal Hilbert-transform and undersampled Hilberttransform. Special attention must be directed to the offset and range of the phase angle. The density distribution of the phase angle is a helpful tool to compensate for offset. Extended Kalman ﬁlters are best suited for further signal processing, i.e., the development of cross-correlation functions. Even analog methods such as the electronic Hilbert-transform or the multiplication of the modulated signal with the carrier signal, lead to very good results for the determination of the ﬂow velocity.

94

5 Signal Processing of Complex Modulated Ultrasonic Signals

References Bachmann W (1992) Signalanalyse. Vieweg Verlag, Braunschweig Filips Ch, Hans V (2002) Comparison of analog and digital demodulation methods of modulated ultrasonic signals in vortex ﬂow metering. IEEE Conf. IMTC 2002, Proceedings:1657–1660 Filips Ch (2003) Ultraschallsignalverarbeitung bei Korrelations- und Vortexverfahren zur Durchﬂußmessung. Dissertation, Universität Essen, also published by Cuvillier Verlag Göttingen (ISBN 3-89873-711-X) Haykin S (1996) Adaptive Filter Theory. Prentice Hall Information and System Sciences Series Lighthill J (1972) The propagation of sound through moving ﬂuids. Journal of Sound and Vibration, 24(4):471–492 Lin Y (2003) Analog signal processing in ﬂow measurement. Techn. Report, Universität Essen, Mess- und Regelungstechnik Loffeld O (1990) Estimationstheorie II, Anwendungen Kalmann-Filter. Oldenbourg Verlag, München Mäusle R (1988) Analoge Modulationsverfahren. Hüthig Verlag, Heidelberg Niemann M, Hans V (2002) New signal processing methods of ultrasound phase recontruction for measurement of disturbed ﬂow. Systemics, Cybernetics and Informatics Conference Proceedings, Vol. III:345–349 Niemann M (2002) Signalverarbeitung in der Ultraschall-Durchﬂussmessung. Dissertation, Universität Essen, also published by Shaker Verlag Aachen (ISBN 3-8322-0340-0) Otnes RK, Enochson L (1978) Applied Time Series Analysis. Wiley, New York/USA Poppen G (1997) Durchﬂußmessung auf der Basis kreuzkorrelierter Ultraschallsignale. Dissertation, Universität Essen, also published by Shaker Verlag Aachen (ISBN 3-8265-31086) Skwarek V (2000) Verarbeitung modulierter Ultraschallsignale in Ein- und Mehrpfadanordnungen bei der korrelativen Durchﬂußmessung. Dissertation, Universität Essen; also published by Shaker Verlag Aachen (ISBN 3-8265-7589-X) Skwarek H, Hans V (2000) The ultrasonic cross-correlation ﬂow meters - new insights about the physical background. FLOMEKO 2000, Proceedings Skwarek H, Hans V (2001) An improved method for hardware-based complex demodulation. Measurement 28/2:87–93 Tribolet JM (1977) A new phase unwrapping algorithm. IEEE Trans. on Acoustics, Speech and Signal Processing 25 (2):170–177 Unbehauen R (1997) Systemtheorie. Oldenbourg Verlag, München Windorfer H (2001) Optimierung von Wirbelfrequenzmeßgeräten mit demodulierten Ultraschallsignalen. Dissertation, Universität Essen; also published by Shaker Verlag Aachen (ISBN 3-8265-9261-1)

6 Vortex-Shedding Flow Metering Using Ultrasound Volker Hans1 , Ernst von Lavante2 1 2

Institut für Mess- und Regelungstechnik, Universität Essen, 45117 Essen, Germany Institut für Strömungsmaschinen, Universität Essen, 45117 Essen, Germany

The coincidence of vortices generated by a bluff body in a gaseous ﬂow, which is well-known as the Kármán vortex street, with an ultrasonic beam crossing these vortices raises many questions concerning physics and signal processing. Usually the vortices are detected by pressure sensors in the pipe wall or inside the bluff body. Using ultrasound for vortex detection is a powerful advantageous alternative combination. The ultrasonic beam is modulated by the velocity components of the vortices and varying ﬂuid density. The kind of modulation depends on the structure of the vortices that are inﬂuenced by the bluff-body geometry and size. Measurements and simulations show that conventional triangular bluff bodies with the large size used for pressure measurements are unsuitable for ultrasonic measurements because of the lack of knowledge about secondary vortices. They can be avoided by turning the bluff body so as to face the tip to the inﬂow. As ultrasound is very sensitive to the smallest inﬂuences, the bluff-body size can be reduced considerably. Even the shape can be adapted. The best results are obtained with a 3-mm threaded rod as the bluff body in a pipe of 100 mm diameter. Pressure losses behind such small bluff bodies can be neglected. Independent of the bluff-bodies shapes and sizes vortex meters are sensitive to disturbed proﬁles and pulsation.

6.1 Introduction Commercial vortex ﬂow meters are based on the well-known relationship between the vortex frequency in the wake of a bluff body and the mean ﬂow velocity. Usually, piezoelectric or piezoresistive pressure sensors, strain gauges or thermal sensors are applied for detecting the vortex frequency. Pressure sensors are installed in the wall of the pipe or inside the bluff body for the detection of separating vortices. Various bluff-body shapes have been designed creating regular and well-deﬁned vortex structure and pressure signals at the sensor. Large bluff-body dimensions are required for strong pressure ﬂuctuations. A width of 24 to 28% of the pipe diameter is recommended in the literature (Baker, 2000; Bentley, 1983; Breier and Gatzmanga, 1998)

96

6 Vortex-Shedding Flow Metering Using Ultrasound

covering an area of more than 30% of the pipe proﬁle and causing immense pressure losses. Signal processing presupposes well-deﬁned vortices at only one dominant frequency. Often the triangular bluff bodies used do not fulﬁl this condition. They generate secondary vortices that are not so well-known, complicating the signal processing. Recent research has studied vortex ﬂow meters that detect the vortex frequency as a modulation of ultrasonic signals (Filips, 2003; Perpeet, 2000; Windorfer, 2001). Special analog and digital signal procedures have been developed for the demodulation of the ultrasonic signal, as shown in Chap. 5. Ultrasonic signals show a higher sensitivity for the detection of vortices. Therefore much smaller bluff-body sizes can be realised with negligible pressure losses. Various bluff-body geometries and kinds of installation have been investigated that were assisted by numerical simulations. This chapter is a summary of research work described in detail in (Filips, 2003; Perpeet, 2000; Poppen, 1997; Windorfer, 2001).

6.2 Physical Background The interaction of streaming ﬂuid and its vortices with an ultrasonic wave results in a complex modulation of amplitude and phase of the ultrasound signal (Poppen, 1997). Different physical effects in the ﬂuid lead to different kinds of modulation (Windorfer, 2001). The ultrasonic wave is damped by acoustic absorption as a consequence of the conversion of mechanical into thermal energy. The intensity of sound varies on changing temperature, density and pressure. As a vortex structure passing an ultrasonic beam represents an area of lower pressure and density than the surrounding ﬂuid, the ultrasonic wave is modulated in amplitude. Each vortex causes one minimum in the time signal of the transmitted amplitude. Experiments and simulations have shown that the modulation degree is about 0.4%. The vortex structures also represent areas of different refraction indices. They are inﬂuenced by the same physical parameters as temperature, density and pressure. Refraction changes the direction of ultrasound propagation and results in interference of the sound. The amplitude is modulated and attenuated. Experimental investigations and measurements have shown a refraction angle of about ±3 ◦ , resulting in an amplitude modulation of about ± 10%. Finally, the ultrasonic beam is deﬂected and elongated by high axial velocity components of the ﬂuid from its perpendicular propagation direction. Because of the cone characteristics, sound waves with lower amplitude are transmitted to the receiver. The drift angle is dependent on the ﬂow velocity and could be found in measurements of about 5% at a ﬂow velocity of 30 m/s. The resulting modulation degree of the amplitude was about 2%. The change of the parameters mentioned above further varies the local sound speed, resulting in phase shift. It could be found that each vortex generates one complete oscillation phase angle with an amplitude of 20%. The phase modulation additionally is caused by the superposition of ﬂow-velocity vectors and sound-speed

6.4 Simulations

97

vectors. The phase modulation can reach values of some hundreds of degrees depending on the ﬂow velocity and the size of vortex structures, which in turn depend on the geometry of the bluff body.

6.3 Measurement Principle and Test Arrangement Vortex ﬂow meters use the well-known effect of a Kármán vortex street. Vortices separate periodically from both sides on the back of a bluff body that is ﬂowed around. Special shaping of the bluff body has an inﬂuence on the linearity of the characteristic of vortex frequency f versus mean ﬂow velocity u m expressed by the dimensionless Strouhal number fd , (6.1) Sr = um where d is the height of the bluff body. If the Strouhal number can be kept constant over the interest velocity range of the mean ﬂow velocity can be determined directly by measurement of the vortex frequency. All measurements were performed in a test arrangement with a pipe diameter of 100 mm. The length of the pipe from the inﬂow to the test chamber with bluff bodies was 5 m to ensure a fully developed ﬂow proﬁle. The ﬂow velocity was controlled in a range 2–30 m/s corresponding to Reynolds numbers 13000–195000 with a deviation of max. 2 cm/s. A turbine gasmeter was used as reference with a deviation within 1% of the mean ﬂow. The test ﬂuid for all experiments was 1 bar static pressure. The ultrasound transducers were usually operated with a carrier frequency of 220 kHz. The location of the ultrasonic beam was selected by the demand for the most sinusoidal time signal of the demodulated amplitude in a range of 10–80 mm behind the bluff body. Various bluff-body shapes were tested, such as triangular, rectangular, T-shape, special developed shapes, circular and threaded control rods. Only the most interesting sizes and shapes will be presented in the following. Signal processing of the complex modulated ultrasonic signals has been performed as described in Chap. 5.

6.4 Simulations The numerical algorithm employed uses the three-dimensional, time-dependent full Navier–Stokes equations describing the conservation of mass, momentum and energy of the ﬂow (Perpeet, 2000). The program is based on the ﬁnite-volume formulation, using a cell-centered organisation of the control volumes. The spatial discretisation is carried out with the help of Roe’s ﬂux difference scheme, a Godunov-type method providing an approximate solution of the Riemann problem on the cell interfaces. The method has been demonstrated to be very accurate and effective in the simulation of low Mach number viscous ﬂows. Upwind-biased differences are used for the convective terms, central differences for the viscous ﬂuxes.

98

6 Vortex-Shedding Flow Metering Using Ultrasound

Starting with a constant initialisation of the scalar variables and body-ﬁtted velocity components, the integration in time is carried out by a modiﬁed explicit Runge–Kutta time stepping as well as, optionally, an implicit approximatefactorisation method (AF) or symmetric Gauss–Seidel (SGS) scheme. The simulations were carried out on structured grids. The mesh points were arranged according to an algebraic distribution and clustered at the solid walls to ensure enough gridpoints in the boundary layers. The domain was divided into several blocks to make the formulation of the boundary conditions and handling a complex geometry easier. Furthermore, the multiblock structure was necessary to compute the ﬂow on parallel computers. For the inlet and outlet planes, the subsonic onedimensional nonreﬂecting boundary conditions were implemented. They were based on the Riemann invariants normal to the pipe cross section. These conditions made the pressure waves and other disturbances run out of the domain without reﬂection. The interzonal boundary conditions provided the data exchange among the blocks; at the walls, the no-slip solid viscous wall boundary conditions were used.

6.5 Comparisons of Pressure and Ultrasonic Signals For basic investigations a conventional triangular bluff body of 24 mm height and 48 mm length was used. The pressure signal was measured with a miniaturised probe of only 1.6 mm diameter 30 mm downstream behind the tip of the bluff body and 20 mm above the centerline of the bluff body. The ultrasonic barrier was also 30 mm behind the bluff body. Both signals were measured at a ﬂow velocity of 15 m/s. Figure 6.1 shows in the upper picture the averaged signal of the demodulated amplitude of the ultrasonic signal in the time domain. In the lower picture the pressure signal is presented. Two points are remarkable. The ultrasonic signal shows in a

Fig. 6.1. Top: Time signal of the demodulated amplitude of the ultrasonic signal. Bottom: Time signal of pressure.

6.6 Experiments

99

higher resolution the inﬂuence of secondary and stochastic effects. Secondly, the frequency of the ultrasonic signal is twice as high as the measured pressure signal. This is effected by the two different measurement principles. The ultrasonic signal detects along the beam, through the whole pipe diameter, all structures in the ﬂuid, especially vortices both in the upper and lower half of the pipe, whereas the pressure sensor measures only the pressure ﬂuctuations passing the measurement location. The same effect is measured with a pressure sensor in the pipe wall. The signal is generated only by vortices in the upper half of the pipe wall. Due to the low sensitivity of pressure sensors in comparison to ultrasound, secondary and stochastic effects are not detected. This means that vortex measurements with ultrasound from the beginning have double sensitivity in comparison to arrangements with a pressure sensor (Hans et al., 1997).

6.6 Experiments 6.6.1 Large Triangular Bluff Body In Fig. 6.2 the simulated pressure plot of a vortex street behind a triangular bluff body is shown (Hans et al., 1998, 2000). The pressure plot was selected for the presentation of the simulation results because of the best visualisation. The velocity ﬁeld circulates clockwise and anticlockwise around the pressure structures. The vortex at the lower side has fully separated but a secondary, smaller vortex develops on the same side. This secondary vortex leads to a secondary maximum in the simulated time signal (Fig. 6.3). The same secondary effect can be observed at the measured phase shift of the ultrasonic signal. It can be readily detected in the frequency-domain measurements. The pressure signal measured in the tube wall is not affected by the secondary vortex (Fig. 6.4). Vortex detection by a pressure sensor shows a very good sinusoidal signal in the time domain. It is superposed by noise. The determination of the signal frequency is very easy in the frequency domain. The vortex frequency is signiﬁcant and overtops the noise with a high signal-to-noise ratio.

Fig. 6.2. Pressure ﬁeld of a simulated vortex street of a triangular bluff body with a width of 24 mm used in the usual way.

100

6 Vortex-Shedding Flow Metering Using Ultrasound Simulated

Amplitude

Measured

Secondary Vortex Time

Fig. 6.3. Measured ultrasonic phase shift and simulated time signal of a triangular bluff body used conventionally.

The shape of the triangular bluff body facing the ﬂat side to the inﬂow is optimised for many commercial vortex-shedding ﬂow meters using pressure sensors for the detection of the vortex frequency. The same shape used for the measurement system combined with ultrasound leads to worse signals. Secondary effects on the ultrasonic signal prevent simple digital signal processing. Using the same shape rotated for vortex generation, measurement and simulation lead, as expected, to very different results with an advantage for the ultrasound measurement method. Figure 6.5 shows a pressure plot of the triangular bluff body facing the edge to the inﬂow (Hans et al., 1998). The vortices separate at the backside without the development of secondary vortices. The vortex street is fully developed in the middle of the pipe. The pressure ﬂuctuations at the wall are very small, so that a detection of the vortex frequency by pressure sensors becomes impossible. In the simulated time signal there is no secondary effect visible. The measured time signal of the demodulated ultrasound signal also shows very sinusoidal behavior without the inﬂuence of

[b] Fig. 6.4. Pressure signal in time (left) and frequency domain (right) of a 24 mm triangular bluff body.

6.6 Experiments

101

Fig. 6.5. Pressure ﬁeld of a simulated vortex street of a triangular bluff body with a width of 24 mm facing the edge to the inﬂow.

secondary effects, but the amplitude varies strongly (Fig. 6.6). The spectrum shows only the smallest secondary effects. The use of the triangular bluff body in the two different directions shows the inﬂuence of ultrasound and pressure sensors on both detection methods. The pressure sensor requires a strong pressure signal, and secondary effects have no inﬂuence on the signal. An ultrasonic barrier is much more sensitive to secondary vortices, it requires well-deﬁned structures without any secondary effects. Not only the time signals and spectra of the two different arrangements of bluff bodies are different, but also the characteristics, as shown in Fig. 6.7. The sensitivity in the conventional arrangement with the ﬂat side to the inﬂow is 6.7 Hz per m/s, corresponding to 6.7 primary vortices per meter. Each primary vortex occupies a distance of 15 cm. The Strouhal number is S r = 0.16. On turning the tip of the bluff body to the inﬂow, the sensitivity more than doubles to 14 Hz per m/s. The distance between successive vortices is 7.1 cm and the Strouhal number is 0.34. It is evident that secondary vortices lead to decreasing sensitivity. For ultrasonic measurements the conventional arrangement of triangular bluff bodies should be avoided. Simulated

Time

Time

Amplitude

Measured

Fig. 6.6. Measured and simulated time signal and frequency spectrum of the ultrasonic phase shift of a triangular bluff body facing the edge to the inﬂow.

102

6 Vortex-Shedding Flow Metering Using Ultrasound 800

Vortex Frequency [Hz]

Vortex Frequency [Hz]

800

600

400

200

0

10

20

30

M ean F low V e lo city [m /s]

40

600

400

200

0

10

20

30

M ean F low V elocity [m /s]

40

Fig. 6.7. Vortex frequency versus mean ﬂow velocity of conventional arrangement (left) and turned around bluff body (right) of 24 mm height with ultrasonic measurement.

6.6.2 Small Triangular Bluff Body Large bluff bodies may be good for vortex detection with pressure sensors, but because of the much higher sensitivity of ultrasonic waves for disturbances it is selfevident to investigate smaller bluff bodies. In the ﬁrst instance, the triangular shape will be maintained. Experiments have shown that the smallest dimension with best results is a bluff body of 4 mm height and 8 mm length. This is a drastic reduction to 4% of the pipe diameter.

[b] Fig. 6.8. Signal in time and frequency domains for a 4-mm bluff body in conventional arrangement.

Simulations and measurements have shown for a conventional arrangement that secondary vortices are generated at the tip of the bluff body following the main vortex and resulting in a second, small amplitude of double frequency in the frequency domain (Fig. 6.8). Physically, it is the same effect as at a large bluff body. The main frequency can be detected deﬁnitely. The 4-mm bluff body shows a much higher sensitivity of 25 Hz per m/s with a corresponding Strouhal number of S r = 0.1. It is most remarkable that pressure losses behind the bluff body decrease substantially by a factor of about 5. On rotating the small triangular bluff body by 180◦ so that the gas ﬂows against the tip of the triangle it can be observed that secondary vortices dissolve. Even here

6.6 Experiments

103

Fig. 6.9. Ultrasound signal of “turned-around” 4-mm triangular bluff body in time and frequency domain.

Vortex Frequency [Hz]

Vortex Frequency [Hz]

the same effect occurs as at the large bluff body. In the time domain they are not recognizable, with only a small effect observed in the frequency domain (Fig. 6.9). This kind of arrangement has an additional advantage. The sensitivity increases to 60 Hz per m/s and is more than twice as high as in the conventional arrangement with the same very small pressure losses. The Strouhal number is 0.24. Figure 6.10 shows a comparison of the characteristics of conventional and “turned-around” arrangements.

Mean Flow Velocity [m/s]

Mean Flow Velocity [m/s]

Fig. 6.10. Vortex frequency versus mean ﬂow velocity of conventional arrangement (left) and a turned-around bluff body (right) of 4 mm height with ultrasonic measurement.

6.6.3 T-Shaped Bluff Body Another type of bluff body used in commercial vortex-shedding ﬂow meters combined with pressure sensors is the T-shape form. For ultrasonic measurements the width of the body can be reduced to only 10 mm, while maintaining signal quality. Figure 6.11 shows the density plot of a T-shaped bluff body used in the conventional way facing the ﬂat side to the inﬂow. The picture is zoomed into the central part of the pipe for better visualisation. The simulation was performed for the whole diameter, but the ﬂow near the wall is hardly affected by the vortices. Also, parallel to the primary vortex a secondary vortex of very low amplitude is separating. The

104

6 Vortex-Shedding Flow Metering Using Ultrasound

Fig. 6.11. Density plot of a simulated vortex street of a T-shaped bluff body with a width of 10 mm used in the conventional way. Only the central part of the pipe is displayed.

measured time signal of the phase shift of the ultrasonic signal shows that the inﬂuence of the secondary vortex is similar to the signal of the triangular bluff body (Fig. 6.12). The simulated time signal is not affected by the secondary vortex. The small inﬂuence can be explained by the simpliﬁcation of the ultrasonic transmission. Measured

Simulated

Amplitude

Secondary Vortex

Time

Fig. 6.12. Measured and simulated time signal of the ultrasonic phase shift of a T-shaped bluff body used in the conventional way.

As for the triangular bluff body, the T-shaped form generates well-deﬁned signals for the ultrasonic detection method if it is used the other way round. The density plot (Fig. 6.13) shows only primary vortices with strong restriction. The simulated and the measured time signals (Fig. 6.14) show a very well-deﬁned sinusoidal behavior that is not disturbed by any secondary vortex. The amplitude modulation of the phase shift is much smaller than for the triangular bluff body. A pressure signal at the pipe wall could not be measured because of the very small dimensions of the body. On comparing the sensitivity of the two arrangements, 12 Hz per m/s could be detected for the conventional arrangement with the ﬂat side to the inﬂow, while the turned-around arrangement resulted in 20 Hz per m/s for the 10-mm bluff body. The corresponding Strouhal numbers are Sr = 0.12 and Sr = 0.2, respectively.

6.6 Experiments

105

Fig. 6.13. Density plot of a simulated vortex street of a T-shaped bluff body with a width of 10 mm facing the bar to the inﬂow. The central part of the pipe only is displayed. Simulated

Time

Time

Amplitude

Measured

Fig. 6.14. Measured and simulated time signal of the ultrasonic phase shift of a T-shaped bluff body facing the bar to the inﬂow.

6.6.4 Circular Form In order to decrease pressure losses, a circular form of only 3 mm diameter has been tested. Figure 6.15 shows the result in the time and frequency domains.

Fig. 6.15. Demodulated ultrasound signal in the time and frequency domains of a 3-mm circular bluff body.

106

6 Vortex-Shedding Flow Metering Using Ultrasound

The main frequency to be detected is fully developed, but amplitudes vary over a wide range because the point of separation of vortices on the round and smooth surface is not stable. Nevertheless, the Strouhal number is Sr = 0.22 and the sensitivity reaches 73 Hz per m/s. This is more than 10 times that of a triangular form in the conventional arrangement. Because of the mentioned instabilities of the separation points of vortices, rough surfaces were supposed to lead to better stability. 6.6.5 Threaded Control Rod Best results with very good sinusodial signals could be obtained with threaded control rods. Figure 6.16 represents the characteristics of threaded control rods of different sizes. 2000 1800

Vortex frequency [Hz]

1600

M4

1400 1200

M5

1000

M6

800

M8

600

M10

400 200 0

0

5

10

15 Velocity [m/s]

20

25

30

Fig. 6.16. Characteristics of different bluff-body sizes of threaded control rods.

It is evident that the sensitivity of the vortex meter depends on the bluff-body size. The sensitivity is given by E=

df in Hz per m/s . dv

(6.2)

This refers to the number of vortices per meter and represents the periodic length of vortices. The sensitivity as a function of the bluff-body height is shown in Fig. 6.17 (Filips, 2003; Hans, 2003). The mean characteristic can be determined by curve ﬁtting as E = 208d −0.93 , with d in mm (Filips, 2003). This very simple relation permits the determination of the vortex meter characteristic as a function of the dimension of the bluff body. Measurements have shown that similar relations are applicable to other bluff-body sizes, too. The best results of various bluff-body shapes were obtained using a M3 threaded control rod. The generated vortices are very stable and no secondary effects could be observed. The frequency is distinct with a good signal-to-noise ratio and lower

6.6 Experiments

107

90

M3

80

Sensitivity [m^-1]

70

M4

60

50

M5 M6

40

M8

30

M10 M12

20

10

M14

M20

M16

M24

M30

0 0

5

10

15

20

25

30

35

Diameter d [mm]

Fig. 6.17. Sensitivity versus size of threaded control rod for undisturbed ﬂow.

frequencies can not be detected (Fig. 6.18). The rough surface of the bluff body increases the friction on the bluff-body surface and clearly deﬁnes the separating point of vortices. The sensitivity increases to 80 Hz per m/s with a Strouhal number of S r = 0.24. Pressure losses decrease to a negligible minimum. The uniform vortex street leads to reliable and sinusoidal demodulated ultrasonic signals that can readily be handled in signal processing. The reduction of the bluff body from 24 mm to 3 mm leads to an enlargement of the cross-sectional area in the pipes resulting in drastically decreasing pressure losses.

Fig. 6.18. Demodulated ultrasound signal in time and frequency domains of a 3-mm threaded control rod.

6.6.6 Pressure Losses Figure 6.19 shows the pressure losses for different bluff-body sizes. The classical triangular size of 24% bluff-body height related to the pipe diameter causes rapidly in-

108

6 Vortex-Shedding Flow Metering Using Ultrasound

creasing pressure losses with increasing ﬂow velocity. By minimization of the bluffbody size pressure losses can be reduced to more than one tenth. Pressure losses of threaded control rod M3 are only negligibly higher than those in a free pipe in the same test arrangement (Windorfer, 2001).

Pressure loss [Pa]

120

Delta Form 24 x 48 mm Rectangular Form 10 x 0,9 mm Threaded control rod M3 Free pipe

100 80 60 40 20 0 0

10

20

30

Mean flow velocity [m/s]

Fig. 6.19. Pressure losses of different sizes and arrangements of bluff bodies.

6.7 Inﬂuence of Disturbances 6.7.1 Single and Double Elbows In industrial applications the necessary requirements of installation for a fully developed proﬁle can not be met in any case. Single and double elbows inﬂuence the velocity proﬁles of the ﬂow. The unsymmetrical velocity distribution behind a single elbow is based on centrifugal forces displacing the maximum of the velocity in the

Single elbow Double elbow

M5

160

Sensitivity in m

-1

M6 120

M8 M10

80

M12

M14

M16 M20

40

0

0

5

10

15

Diameter in mm

20

25

Fig. 6.20. Sensivity versus size of threaded control rod for disturbed ﬂow.

6.8 Conclusion

109

outer range of the pipe. Additionally the main ﬂow is superposed by a cross-ﬂow resulting in a double vortex. Double elbows, however, result in a ﬂow proﬁle with spin. Investigations on the inﬂuence of unsymmetrical and spin-affected ﬂow have been made with threaded control rods M3 to M30 in a distance of the double pipe diameter 2D. Sizes smaller, M5, and larger, M20, are too sensitive on disturbed proﬁles, especially behind a single elbow (Filips, 2003). It is remarkable that the vortex frequency is nearly twice as high as for the undisturbed ﬂow. Figure 6.20 shows the sensitivity as a characteristic of the threaded-rod size for disturbed ﬂow. Measurements behind a single elbow show high deviations independent of the kind of signal processing. Obviously they must be explained with the physics of vortex generation. Measurements behind a double elbow, however, lead to more reliable results. The vortex street is more stable. Nevertheless, the deviations are higher than for the undisturbed proﬁle. 6.7.2 Pulsation Two types of pulsation generators have been applied to the vortex meters. The ﬁrst was an industrial lobed-impeller ﬂow meter (lif) generating a discontinuous ﬂow of a deﬁned volume by the rotating piston. The second was a rotating plate inside the pipe with a plate diameter of 0.7 D (D = pipe diameter). Measurements have been made with triangular bluff bodies and threaded control rods in a distance of 2 D, 20 D and 40 D to the pulsation generator. For the lobedimpeller ﬂow meter, in general, it can be stated that the energy of the pulsating ﬂow is higher than the energy of vortices generated by bluff bodies. Therefore, mostly the pulsating frequency was determined with high deviations independent of the bluffbody shapes and sizes even in a distance of 40 D. The pulsation frequency of a rotating plate inside the pipe could be adjusted independent of the ﬂow velocity. A lock-in effect could be detected only if the natural vortex frequency was very close to the adjusted pulsating frequency. The turnedaround triangular bluff body with the tip to the inﬂow was less sensitive to pulsation effects than the conventional bluff-body arrangement with the ﬂat side to the inﬂow. All measurements have shown large deviations. It can be stated, in general, that vortex ﬂow meters are very sensitive to any kinds of pulsation.

6.8 Conclusion The application of ultrasound in vortex-shedding ﬂow meters is a powerful combination. The ultrasonic signal is modulated in amplitude and in phase by the vortices. The demodulated amplitude signal has been evaluated. Compared with conventionally used pressure sensors, the ultrasonic method shows higher sensitivity. This

110

6 Vortex-Shedding Flow Metering Using Ultrasound

method requires a new design for measuring systems. Simple signal processing presupposes sinusoidal signals that can be generated by triangular bluff bodies rotated 180◦ compared with conventional arrangements. The size can be drastically reduced to 4% of the pipe diameter, leading to negligible pressure losses behind the bluff body. The best results were obtained with an M3 threaded control rod with a sensitivity of more than ten times that of conventional, pressure-based measurements. Independent of the bluff-body shapes and sizes ultrasonic vortex meters are sensitive to disturbed ﬂow proﬁles and pulsations.

References Baker RC (2000) Flow Measurement Handbook. Cambridge University Press Bentley JP (1983) Principles of Measurement Systems. Longman, London and New York Breier A, Gatzmanga H (1998) Parameterabhängigkeit der Durchﬂuss-Frequenz-Kennlinie von Vortex Zählern im Bereich kleiner Reynolds Zahlen. Technisches Messen 62:22–26 Filips Ch (2003) Ultraschallsignalverarbeitung bei Korrelations- und Vortexverfahren zur Durchﬂußmessung. Dissertation, Universität Essen, also published by Cuvillier Verlag Göttingen (ISBN 3-89873-711-X) Hans V, Poppen G, von Lavante E, Perpeet S (1997) Interaction between vortices and ultrasonic waves in vortex shedding ﬂowmeters. FLUCOME 97, Proceedings Vol. 1:43–46 Hans V, Windorfer H, von Lavante E, Perpeet S (1998) Experimental and numerical optimization of acoustic signals associated with ultrasound measurement of vortex frequencies. FLOMEKO 98, Proceedings:363–367 Hans V, Windorfer H, Perpeet S (2000) Inﬂuence of vortex structures on pressure and ultrasound in vortex ﬂowmeters. IMEKO 2000, TC9, Proceedings Hans V (2003) Detection of vortex frequency in gas ﬂow using ultrasound. Int. Symp. on measurement technology and intelligent instrumentation, Hongkong, Proceedings Perpeet S (2000) Numerische Simulation von Strömungsfeldern um DurchﬂußMeßanordnungen. Dissertation, Universität Essen; also published by Shaker Verlag Aachen (ISBN 3-8265-7219-X) Poppen G (1997) Durchﬂußmessung auf der Basis kreuzkorrelierter Ultraschallsignale. Dissertation, Universität Essen; also published by Shaker Verlag Aachen (ISBN 3-8265-31086) Windorfer H (2001) Optimierung von Wirbelfrequenzmeßgeräten mit demodulierten Ultraschallsignalen. Dissertation, Universität Essen; also published by Shaker Verlag Aachen (ISBN 3-8265-9261-1)

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods Volker Hans Institut für Mess- und Regelungstechnik, Universität Essen, 45117 Essen, Germany

Cross-correlation functions are in widespread use especially for the measurement of wave-propagation time and all kinds of velocity. The coincidence of turbulent structures in a gaseous ﬂow with ultrasonic waves raises many questions concerning physics and signal processing, in particular on account of dissipating structures between the two ultrasound barriers. It appeared that the ﬂow velocity measured by the cross-correlation function of the complex modulated signals is determined by the most frequent velocity components in the ﬂuid. Besides, the ultrasonic beam measures along a line, whereas the average ﬂow velocity is deﬁned by an area integral. This velocity-dependent nonlinear effect necessitates a calibration of the measurement device. Uncertainties better than one percent are obtainable. Even disturbed ﬂow by single and double elbows or pulsation can be measured in a qualiﬁed sense concerning uncertainty and scattering. Multipath arrangements can improve scattering. Furthermore, using special algorithms the proﬁle of disturbed or undisturbed ﬂow can be represented by tomography.

7.1 Introduction In recent years an increasing market of ﬂow-measuring instruments using ultrasound can be observed. Contactless measurements, no mechanical parts in the ﬂow, no pressure drops are evident advantages of this technique. Usually, the propagation time of ultrasound based on the Doppler effect is used to determine the ﬂow velocity (Schrüfer, 1995a). Cross-correlation methods are well known but not so widespread on account of difﬁcult signal analysis. The signals to be correlated are modulated only slightly by the ﬂow and the sidebands are very close to the carrier frequency. In this case, signal processing with correlation functions is very difﬁcult. Especially if the carrier frequency is very high, as for ultrasound in the range of some hundred kHz, oversampling techniques with some megahertz lead to problems in realtime signal processing on account of the limited computer storage, see Chap. 5.

112

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

Transmitters

Flow direction

v

Receivers Fig. 7.1. Cross-correlation measurement device.

Modern principles of signal processing improve the possibilities of research in countless scientiﬁc areas. In this chapter, a practical application of such methods is presented for the explanation and the improvement of an ultrasound cross-correlation ﬂow meter. It is used for the determination of the volume ﬂow, which is proportional to the average axial ﬂow speed. Although the basic theory of this measurement is simple and easy to understand, much effort was necessary to improve the instrument range from multiphase and particle ﬂow to gaseous ﬂuids. The main reason for these difﬁculties is based on the unsatisfactory knowledge of the properties of turbulent ﬂuids and the interaction with ultrasound, see Chap. 6. Therefore, there are still unexplainable systematic deviations and unexpected behavior in the measured data. By performing a reverse processing of that data more information about the ﬂuid can be gained and a further understanding and improvement of this device is possible. This chapter is a summary of research work described in detail in Filips (2003); Niemann (2002); Poppen (1997); Rettich (1999); Skwarek (2000). The idea for this device was developed between 1935 and 1955 by several scientists, but no exact date and origin can be determined. A ﬁrst patent was registered in 1967 with a corresponding PhD thesis by Beck in 1969 with a huge number of publications following by other scientists. For further references see Beck (1987). As a main intention the feedback on the ﬂuid should be reduced by using a continuous ultrasonic wave transmitted between transmitter–receiver pairs in the radial direction perpendicular to the main ﬂow. For multiphase and particle ﬂows, the amplitude of the ultrasonic beam s1 is modulated stochastically with the density and distribution of the particles crossing the beam. Each combination leaves a certain pattern like a ﬁngerprint in the received signal. Assuming that the ﬂuid and the particle distribution do not change signiﬁcantly within a short distance, a second downstream transmitter– receiver pair s2 strongly receives the same signal – only with a time delay according to the traveling time of the particles between the ultrasonic barriers (Fig. 7.1). The point of maximum similarity between s1 and s2 is computed by using the mathematical cross-correlation function (CCF) Φ12 1 Φ12 (τ ) = lim T →∞ 2T

T s1 (t)s2 (t − τ )dt −T

(7.1)

7.2 Determination of Traveling Time by Cross-Correlation Functions

113

and the result can be directly read from the abscissa of the extremum of the CCF (Fig. 7.2). With a known distance d between the barriers the average axial speed can be calculated by d (7.2) v¯ = . t

Fig. 7.2. Normalized result of the CCF between s1 and s2 . Unique maximum at t = 0.01 s represents traveling speed of structures v¯ = 10 m/s.

7.2 Determination of Traveling Time by Cross-Correlation Functions For digital signal processing (7.1) the cross-correlation function can be written as N 1 s1 (nTa )s2 ((n − k)Ta ) , Φ12 (kTa ) = T n=1

Φ12 (kTa ) =

N 1 s1 (nTa )s2 ((n − i − k)Ta ) , T n=1

(7.3)

Φ12 (kTa ) = Φ11 ((k + i)Ta ) , with the sampling interval Ta , N number of samples and n sampling index. Further details of signal processing are described in Chap. 5. The cross-correlation function equals the autocorrelation function Φ11 shifted by the traveling time iTa (Schrüfer, 1995b). As the magnitude of CCF depends on the root mean square of single-signal components usually the absolute value of the normalized CCF is applied

114

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

Φ12 (kTa ) . |Φ12norm (kTa )| = ' Φ11 (0)Φ22 (0)

(7.4)

The magnitude ranges between 0 and 1 as a measure of the similarity of both signals. Only signals with the same frequency can be correlated. On account of dissipation the similarity of the two signals decreases. The degree of similarity depends on the distance between the two ultrasound barriers. The exact determination of the maximum of CCF and the traveling time depends on the relative deviation given by the sampling frequency fa and distance (Niemann, 2002), v ∆v = . v dfa − v

(7.5)

For a given sampling frequency small distances cause greater deviations especially for high velocities. Small deviations require long barrier distances leading to higher dissipation. Best results could be obtained with distances of 60 to 80% of the pipe diameter D.

7.3 Physical Quantity Measured by Cross-Correlation Functions Many measurements have shown deviations between the measured ﬂow velocity by CCF and the real mean ﬂow velocity. As explained in Chap. 6 various physical effects regarding the interaction between the ﬂuid and the ultrasonic wave cause amplitude and phase modulation of the ultrasound signal. The question arises if both kinds of modulation result in the same measured time, whereas the causal connections are different. The next question is which physical quantity of turbulent structures is measured in the correlation measurement. Many authors gave various answers (Braun, 1984; Niemann, 2002; Poppen, 1997; Schneider, 2001; Shu, 1987; Skwarek, 2000; Worch, 1998) developing interesting models. Interesting considerations conﬁrming the systems theory model in Chap. 8 have been presented by Niemann (2002). Both ultrasonic barriers register many turbulent structures with different velocities in each case. The amplitude- and phase-demodulated signals therefore are composed by single-signal components with different traveling times. The question is which of these components determines the maximum of the correlation function. In a ﬁrst step it may be supposed that both signals s1 and s2 are composed by two components s1 = sx1 (nTa ) + sy1 (nTa ) , s2 = sx2 (nTa ) + sy2 (nTa ) ,

(7.6)

with different traveling times t1 = i1 Ta and t2 = i2 Ta . The CCF yields Φ12 (kTa ) =

N 1 [sx1 (nTa ) + sy1 (nTa )] [sx2 ((n − k)Ta ) + sy2 ((n − k)Ta )] , N n=1

Φ12 (kTa ) = Φx1x2 (kTa ) + Φx1y2 (kTa ) + Φy1x2 (kTa ) + Φy1y2 (kTa ) .

(7.7)

7.3 Physical Quantity Measured by Cross-Correlation Functions

115

The components sx1 (nTa ) and sy1 (nTa ) are supposed to be stochastically independent of each other resulting in Φx1y2 (kTa ) and Φy1x2 (kTa ) equal zero: Φ12 (kTa ) = Φx1x2 (kTa ) + Φy1y2 (kTa ) .

(7.8)

The CCF of both signals corresponds to the sum of both CCF of the single-signal components. Both components generate their own maximum superposing each other. The traveling time is to be determined by the highest maximum. The traveling time of the other component does not inﬂuence the result. If both maxima are close to each other both correlation functions are superposed to one characteristic. The position of this new maximum is not given by the arithmetic mean value of the position of the single maxima. From two components the general case with m signal components can be expressed as m Φxlyl (kTa ) . (7.9) Φ12 (kTa ) = l=1

The CCF of the sum of signals results in the sum of single CCF of all signal components. This means that without dissipation the most frequent velocity component will be detected. The bell-shaped characteristics differ clearly in magnitude as well as in their width. The magnitude is determined by similarity and dissipation of components, while the width represents the bandwidth of the signals. Narrowband signal components generate wide shapes of CCF, wideband components lead to narrow shapes of CCF. Accordingly, four different stochastically independent inﬂuences determine the position of the maximum of CCF: • The number of recognized signal components • The magnitude of amplitudes of signal components • The width of summarized characteristics and thus the bandwidth of signal components • The similarity of components in both barriers Summing up, it may be said that by evaluating the CCF the most frequent velocity component will be measured generating a high degree of modulation of the carrier signal with the maximum rate of change and high amplitudes with low dissipation. The measured velocity by cross-correlation corresponds to the modal value of the probability density function (PDF) of velocity components. It does not represent any mean value of all detected velocity components. As pointed out in Chap. 8 measurements of ﬂuid components by means of particle-image velocimetry have shown that the PDF of velocity components is skewed. The modal value of velocity components is given by the maximum of the skewed PDF (Braun, 1984; Schneider, 2001; Skwarek, 2000). Correlation functions presuppose ergodic processes. A stochastic process is called ergodic if its ensemble averages equal appropriate time averages (Papoulis, 1984). As the modal value is a mean value within the meaning of statistics the theorem of ergodicity is fullﬁlled. These reﬂections are in agreement with the analysis by systems theory in Chap. 8.

116

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

7.4 Measurements Experiments have been made in a test rig with a pipe diameter of 100 mm. The stationary turbulent gas ﬂow was controlled by a calibrated turbinemeter for average ﬂow velocities up to 30 m/s. The distance between the ultrasonic barriers was 100 mm, and the carrier frequency was 220 kHz. Signal processing was executed by digital undersampling and Hilbert-transform as described in Chap. 5. As the effect of phase modulation was more distinct than that of amplitude modulation the determination of CCF was related to the phase-demodulated signal. Measurement results are represented as normalized characteristics according to (7.4). The result is calculated in units of Ta . However, with the time antiproportional to the velocity, the graphical results are not directly interpretable due to its nonlinear behavior. Increasing speed shifts the results not to inﬁnity but to zero, and the width of the maximum peak becomes smaller instead of broader. This is a reason for rescaling abscissa. By applying (7.2) on each value of the abscissa the function Φ 12 (kTa ) is converted into Φ12 (v) with equally spaced velocity units, Fig. 7.3 (Skwarek and Hans, 1999).

Fig. 7.3. Rescaling abscissa from time to velocity.

According to physical assumptions, amplitude modulation is caused by diffraction, reﬂecion and damping due to density changes as described in Chap. 6. Phase modulation, on the other hand, is caused by velocity components in the direction of the ultrasound de- and accelerating signal. Both effects are linked. Turbulent structures are a combination of local pressure, density and velocity distribution and move

7.4 Measurements

117

with the average velocity of the ﬂuid. Furthermore, they dissipate due to the turbulent viscosity surrounding the structures and become larger with a decreasing energy density. Although the structures become larger the average velocity distribution stays the same. Nevertheless, the evaluation of amplitude modulation is not suitable for calculating the average ﬂow velocity because the measurement does not refer to the same relative position in a turbulence (Skwarek and Hans, 1999). 7.4.1 Measurement Uncertainty As mentioned above, the modal value of the ﬂow-velocity components is measured instead of the average value. This modal value is dependent on the ﬂow proﬁle and thus on ﬂow velocity. As the relationship between modal value and average value changes, a calibration of cross-correlation ﬂow meter is necessary in any case. The measurement principle itself causes further systematic deviations between the real mean ﬂow velocity and the measured ﬂow velocity by CCF. Unlike the deﬁnition of the average ﬂow velocity by an area integral 1 v(A)dA , (7.10) v¯def = A A

an ultrasonic beam represents a measurement along a line 1 v(L)dL , v¯meas = L

(7.11)

L

with L and A sizes of the line and the area, respectively. The radial velocity distribution for a fully developed, undisturbed ﬂow shows a speed-dependent exponentiallike curvature due to the friction at the wall. This curvature causes a relative deviation of the measurements 1 v¯meas (7.12) =1+ r= v¯def 2n with n ∈ {6, 7, 8...} for increasing v and n = f (v), see Chap. 8. Usually the theoretical values of r are in the range of 1.085 to 1.05 (Nikuradse, 1932). However, practical tests show further deviations between expected and real results necessitating a calibration of the ﬂow meter. The probable reason for this deviation is the difference between the theoretical assumption of the ultrasonic beam as a line. In reality it covers a non-negligible volume of the test rig and measures not only the centerline of velocities but even slower velocities at the left and right of the centerline. Consequently, the real measured velocity is lower than expected. 7.4.2 Measurement of Disturbed Flow Proﬁles Measurements Behind a Single Elbow In industrial applications disturbed proﬁles in consequence of installations such as single or double elbows are often typical and unavoidable. Measurements have been

118

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods 5

Relative deviation [%]

5

0q

D

0 -5

-5

-10

-10

-15 0 5

5

10

20

15

D

0

60q

-15 0 5

-5

-5 -10 5

10

15

20

D 120q

0

5

10

-15 0 5

D

5

10

15

30q

20

15

0

-10 -15 0 5

D

0

90q

20

D 150q

0 -5

-5 -10

-10

-15 0

-15 0

5

10

15

20

5

10

15

20

v [m/s]

Fig. 7.4. Measurement results in ultrasound paths at different angles behind a single elbow.

made in a distance of 2D = 200 mm behind a 90-degree single elbow. In this case a secondary ﬂow is superposed to the main ﬂow leading to a transverse motion in the form of a spiral double eddy. Examining the inﬂuence of the asymmetric proﬁle the ultrasonic path has been rotated by six angles. The results of relative deviations of the average velocity and the standard deviation of measurement are shown in Fig. 7.4 (Niemann, 2002). It is remarkable that all measured velocities are lower than the real velocity independent of the angle. On account of the unknown asymmetric proﬁle no deviation factor r could be determined. To compensate the deviations the measurement device must be calibrated. The results of the calibrated device are shown in Fig. 7.5. The relative deviations are smaller than one percent with standard deviations of about one percent. The angle of the ultrasound path has practically no effect on the uncertainty. Measurement Behind a Double Elbow Two single elbows are distorted to each other by 90 degrees in space. By this means a spin-affected ﬂow is generated. The relative deviations of the uncalibrated device show characteristics from positive to negative values (Niemann, 2002). Best results could be obtained at an angle of 60 degrees. As for the single elbow the results can be improved by calibration. The average values of deviations are in the range of one percent. The standard deviation increases up to two percent with increasing velocity.

7.4 Measurements 5

5

2.5

2.5

0

0

Relative deviation [%]

-2.5 -5

0

5

10

-2.5

0q

D

0 5

5 2.5

2.5

0

0

-2.5 -5

D

0

5

10

15

0 5

2.5

0

0

0

10

15

15

5

10

15

5

10

15

-2.5

D 120q 5

10

-5

20

0

30q 20

D

-5

20

2.5

-5

5

-2.5

60q

5

-2.5

D

-5

20

15

119

90q 20

D 150q 20

v [m/s]

Fig. 7.5. Measurement results with calibrated measuring device behind a single elbow.

Pulsation Effects Possible inﬂuences of pulsation on a ﬂow meter are manifold so that they will not be completely described in this chapter. Alternatively, the main focus will be put on problems occurring in combination with an ultrasonic measurement device (Skwarek et al., 2001). Therefore, effects are generally divided into two subgroups: • Direct effects on transmitter or receiver • Modulation-like effects on ultrasonic wave Direct effects are caused by shock-waves due to pulsation: Piezoelectric receivers are sensitive to every kind of pressure independent of their origin. Every pressure wave from pulsation is superposed onto the modulated ultrasonic signal at the receiver. The following estimation gives an impression about the magnitudes of the pulsation amplitude in comparison with the ultrasonic wave: The transmitters are continuously driven with a sinusoidal supply voltage umax = 10 V. According to the datasheets of the sensor (Massa, 1999) a transmitting sensitivity of 20 dB versus 1 µbar per Volt has to be expected, which is equivalent to a maximum pressure of 10 Pa. Assuming a wave propagation without any damping losses these 10 Pa are superposed by pulsation. With the fans producing a pressure drop of about 6000 Pa in the pipe, theoretically a pressure pulse of the same magnitude appears in a pipe opened and closed for a short time. Consequently, ultrasonic signals are hardly detectable within the pulsation for disadvantageous boundary conditions concerning pulse frequency and speed of ﬂow. An example of such poor

120

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

Fig. 7.6. Example of the superposition of the ultrasonic signal with modulated turbulence noise and pulsation.

conditions is given in Fig. 7.6. It is easy to imagine that cross-correlation as well as vortex ﬂow meters using nonselective algorithms react sensitively to this pulsation. For pulse amplitudes similar to or less than the modulation amplitude the second case mentioned above has to be taken into consideration. Not only do these impacts directly inﬂuence the sensors, they also inﬂuence the behavior of the ﬂuid. As already explained, modulation properties depend on the boundary conditions given by the environment such as temperature, pressure, density and velocity ﬁelds. According to the magnitude of the pulse one or several quantities change signiﬁcantly. Consequently these changing ﬁelds modulate the sinusoidal carrier as well. Their inﬂuence can possibly be higher than the turbulence-caused modulation itself. For more considerations about the interaction between ultrasound and ﬂuid see Kolpatzik (1999); Poppen (1997); Rettich (1999); Skwarek (2000). Beyond the general effects of pulsation on ultrasonic measurements further inﬂuences concerning the physical principle of a device can be thought of. Coherent structures at the correlation method are unstable and not very powerful. They could either be changed or even destroyed by the pressure, so they cannot be redetected at the second barrier or the pressure moving with sound speed produces such a high amplitude at the receiver that ultrasound cannot be received at all. In industrial applications pulsations are usually generated by pumps, valves or measurement devices such as oval gear meters. For modeling all these disturbances a simple pulse generator was designed that imitated a wide variety of pulses. It consists of an actively driven rotating ﬂap closing and opening the cross section of the pipe sinusoidally with a frequency between 5 and 40 Hz. By using ﬂaps of different diameters and shapes the intensity and type of pulsation can be changed. The complete test rig consists of a straight pipe with a diameter of D = 0.1 m and an

7.4 Measurements

121

overall length of 90.5 D = 9.05 m. After an inlet of 20D the measuring device is installed with the pulsation generator 10D further downstream. A turbine ﬂow meter after 15D is used for reference although this reference gives only an estimation because the ﬂow type exceeds the prescribed setup conditions. An expected, systematic deviation is not corrected yet. Finally, the test rig ends 32D after the reference with the underpressure fan. For air as ﬂuid the mean velocity can be varied between Re = 20000–200, 000 and stabilized on about 99–101% of the desired set point. For the evaluation of the inﬂuence of technical devices and the comparability to the rotating-paddle measurements have been performed with an oval gear meter. In this device the rotating frequency of the oval gear depends linearly on the mean ﬂow velocity so the pulsating frequency is proportional to the mean velocity. The amplitude of the pulsation is conjunct with the volume ﬂow, too. Independent manipulation of the parameters is impossible. For the evaluation of the results the lack of an exact reference has to be taken into consideration. As already mentioned, the turbine ﬂow meter is only suitable for estimating the average ﬂow speed. Therefore, the measurements cannot be evaluated absolutely but only checked for their plausibility. The measurements were performed at velocities v¯ = 5, 10, 15 and 20 m/s with ﬂap diameters varying between 60 mm and 100 mm and a rotating frequency between 0 Hz (open ﬂap) and 40 Hz. Obviously two states of measurements have to be considered: Either all or no measurements are correct for a given boundary condition. Flap diameters up to 80 mm and rotating frequencies up to 20 Hz have no inﬂuence on correct measuring results. The cross-correlation method seems to be very sensitive to pulsation, independent of the average ﬂow velocity. The only parameters deﬁning the behavior of the device are the pulsation frequency and the ﬂap diameter. For further interpretation the ﬂap diameter is set equivalent to the amplitude of the pulse wave because the lower the diameter the higher the amount of bypassing ﬂuid that is not inﬂuenced by pulsation. For understanding the inﬂuence of pulsation amplitudes in detail formula (7.1) for the cross-correlation function of s1 (t) and s2 (t) has to be used. According to the Wiener–Chintschin theorem (Schrüfer, 1995b) this is equivalent to the cross-power spectrum in the frequency domain CPS(ω) =

1 S1 (ω)S2∗ (ω) . 2T

(7.13)

The spectrum of a turbulent ﬂuid contains frequencies from 0 to 3000 Hz with higher amplitudes at lower frequencies (Fig. 7.7). Generally, the power is distributed stochastically over the whole bandwidth. If the ﬂuid starts pulsating with a high amplitude at both sensors for signals s1 and s2 pressure waves move with the speed of sound and appear nearly simultaneously at both ultrasonic barriers. This causes a very powerful periodic component that is squared in the CPS and superposes the maximum caused by the traveling time in the CCF (Fig. 7.8). Although the oscillation of the CCF in the right diagram looks very harmonic even a single bandstop ﬁlter

122

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

Fig. 7.7. Bandwidth of a demodulated ultrasonic signal of a steadily ﬂowing ﬂuid.

does not help to detect the maximum of the CCF. The vibrating ﬂuid makes several coupled systems in the test room resonate with spectral lines that can also be found in Fig. 7.8. Only after ﬁltering the frequency of several maxima and multiples of it up to 200 Hz with bandstop ﬁlters can a sufﬁcient result be obtained. The difference between the ﬁltered and the unﬁltered signal is shown in Fig. 7.9. Here a unique maximum in the CCF can be evaluated. Although ﬁltering seems to be a solution of the problem, two further difﬁculties arise: • The main information of the turbulent ﬂuid is found at frequencies below 200 Hz (Skwarek, 2000; Skwarek and Hans, 1999). Although the remaining, highfrequency part can still be evaluated, the “ﬁltered maximum” differs from the “unﬁltered maximum” by several sampling points. This is equivalent to some percent of the calculated speed. • Furthermore the bandwidth and starting frequency and the number of bandpass ﬁlters is unknown and depends on the velocity of the ﬂuid and pulsation parameters. Lobed-Impeller Gasmeter For the veriﬁcation of the measurements on pulsating ﬂow generated by an artiﬁcial pulsation generator a lobed-impeller gasmeter was also used. The comparison of the two different measurement setups with rotating paddle and lobed-impeller gasmeter shows a similar inﬂuence on the measurement system and the disturbing effects in the frequency domain. The limited independent variation of the pulsation parameters of the second setup shows only a small section of the variety of pulsation parameters. Quantiﬁcation of the pulsation in ﬂow parameters as time-dependent pressure and velocity components should be carried out for a better numerical comparison of different pulsation generators and the unsteady ﬂow.

7.4 Measurements Cross-Power Spectrum

Cross-Correlation Function

Fig. 7.8. Faulty calculated cross-correlation function caused by pulsating ﬂow.

Cross-Power Spectrum

Cross-Correlation Function

bandstop filtered at frequencies of several maximum positions Cross-Power Spectrum

Cross-Correlation Function

Fig. 7.9. Filtered and unﬁltered signals.

123

124

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

7.5 Multipath Arrangement As shown, single-path measurements with ultrasound result in scattering on account of stochastically distributed structures and changes of velocity proﬁle. Furthermore, only a part of the cross section of the pipe is considered by a single ultrasonic beam. Measurements at another position of the pipe increase the chance that further structures are met, equalizing the proﬁle disturbance. Theoretically, best results should be expected by spatial averaging over the cross section of the pipe. This can be approximated with several ultrasonic beams in a multipath arrangement (MPA). First examinations of MPA and problems were introduced in Papoulis (1984) with some basic consideration about the feasibility of measurements and the design of the pipe. With better algorithms for signal processing, this work is particularly being continued. Simplifying assumptions are the existence of a fully developed ﬂow proﬁle and diametric ultrasonic paths with ﬁnite dimensions, crossing exactly in the center of the pipe. For preliminary examinations, a three-path conﬁguration is taken into consideration. Measurements using the three single paths and their common average are shown in Fig. 7.10 (Skwarek and Hans, 2000). The range of the set speed had to be limited for the introductory measurements to 20 m/s. As already mentioned, it is still not 1.12

0.03

path 1 path 2 path 3 average

1.1

rel. scattering

r-factor

1.08

0.025

1.06 1.04 1.02

0.02 0.015 0.01 0.005

1 5

0 15

10

vref [m/s]

20

5

15

10

20

vref [m/s]

Fig. 7.10. Results of a multipath measurement with sensors in star arrangement.

possible to estimate the quality of the r-factor. All three paths as well as their average run in similar intervals, which are at least plausible compared with the expected value. The real improvement of averaging over multiple paths is obvious in the right diagram: The scattering of all three paths shows a different and nonmonotonous behavior, which is signiﬁcantly reduced by their common averaging. This is an indication of oppositely effecting ﬂuctuations at the same time but at different positions in the pipe. Consequently, the mean value over several paths will equalize and suppress these effects.

7.6 Tomographic Reconstruction of Flow Proﬁle

125

7.6 Tomographic Reconstruction of Flow Proﬁle Multipath measurements can be extended into tomographic representations of turbulent ﬂow proﬁles. Only a short overview about ultrasound tomography with crosscorrelation functions and the resultant problems will be given. More details are described in Filips (2001). In the case of disturbed proﬁles, however, simple averaging is not sufﬁcient due to missing rules about weighting the results over the single paths. Therefore, a tomographic representation of the proﬁle could be helpful to determine the total average velocity of the ﬂuid by an integration over this proﬁle. The results of CCF as measures of the average velocities over the ultrasonic paths can be used for a reconstruction of the axial velocity distribution on the ﬂuid. Main problems arise with the deﬁnition of boundary conditions for tomographies. One parallel projection is ideally represented by the entity of an inﬁnite number of parallel beams. This means an inﬁnite number of projections. Parallel beams can be realized in the test rig by a grid arrangement. With respect to crosstalk between the sensors and the macroscopic size of the beams only two perpendicular projections with three beams each are performed. Furthermore, the parallel expansion of the ultrasonic beam as a cone is not ensured. These six paths represent the only information of the ﬂow. They are used for the Radom transformaART

Theoretical flow profile

12

1.2 10

Velocity [m/s]

1

Velocity [m/s]

0.8 0.6

8 6

0.4

4

0.2

2

0 50

0 50 50

50

0

0

0

-50

Radius [mm]

-50

Radius [mm] -50

Radius [mm]

Collocation Kollocation

-50

Interpolation

12

12

10

10

8

Velocity [m/s]

Velocity [m/s]

0

Radius [mm]

6 4 2

8 6 4 2

0 50

0 50 50

50 0 0

Radius [mm]

Radius [mm] -50

0 0

Radius [mm]

Radius [mm]

-50

-50

-50

Fig. 7.11. Fully developed ﬂow proﬁle by different reconstruction algorithms. Upper row left: theoretical proﬁle (Gersten and Herwig, 1992); right: reconstruction by ART. Lower row left: collocation method; right: interpolation method.

126

7 Ultrasonic Gas-Flow Measurement Using Correlation Methods

tion. Special procedures for minimal sets of data must be applied, which are known as the algebraic reconstruction technique (ART). Using the same distribution of pixels as in ART a linear combination of several constant approach functions can be applied, which is known as the collocation method. The tomographic representation can be improved by a simple interpolation method with polynomial functions. For the undisturbed proﬁle a function is taken as a basis that can be changed by matched polynomial functions. The theoretical description of the proﬁle is applied following Gersten and Herwig (1992) as reference for those reference polynomial functions that are to be reached in the reconstruction. The different kinds of tomographic reconstruction are shown in Fig. 7.11 (Filips, 2003, 2001). For the tomographic representation only a limited number of measuring paths with small content of information is available. The collocation method results in a reconstruction with the smallest deviations for the undisturbed proﬁle. This is caused by the rough partition of the cross-sectional area. The irregular reconstruction by ART was obtained by a simpliﬁed algorithm. The reconstruction by interpolation using polynomial functions shows remarkable improvements. The reconstruction methods described have also been applied to disturbed ﬂow after single and double elbows (Filips, 2003, 2001). For the disturbed proﬁle no satisfactory results could be obtained with the collocation method. Best results could be realized with ART showing clearly local maxima and minima.

7.7 Conclusion Cross-correlation ﬂow meters are useful and advantageous tools in gas-ﬂow measurement. Ultrasound is very sensitive to any kind of turbulence in a gas ﬂow. The ultrasonic signal is complex modulated in amplitude and in phase, which may be explained by physical interpretation of the interaction between streaming ﬂuid and the ultrasonic beam. Furthermore, it could be shown that the most frequent velocity components in the ﬂuid determine the magnitude of the cross-correlation function and thus the measured ﬂow velocity. Additionally, the average ﬂow velocity is deﬁned by an area integral over the cross section, whereas an ultrasonic beam represents a line integral. Therefore ultrasonic cross-correlation ﬂow meters have to be calibrated in any case. Uncertainties less than one percent are available. Even for disturbed ﬂow by single or double elbows calibrated ﬂow meters can be installed with uncertainties of one to two percent. Cross-correlation ﬂow meters are very sensitive to pulsation, independently of the average ﬂow velocity. The only parameters deﬁning the behavior of the device are the pulsation frequency and amplitude. Either all or no measurements are correct. Multipath arrangements consider a larger area of the cross section of the pipe and therefore lead to lower scattering. But they also have to be calibrated. They can be used for tomographic representations of the turbulent ﬂow proﬁle. Special procedures for minimal sets of data can be applied to the view of disturbed or undisturbed ﬂow proﬁles.

References

127

References Beck MS (1969) Powder and ﬂuid ﬂow measurement using correlation techniques. PhD thesis, University of Bradford Beck MS (1987) Cross correlation ﬂowmeters: their design and application. IOP Publishers, Bristol Braun H (1984) Statistik der Signale berührungsloser Strömungssensoren. Dissertation, Universität Karlsruhe Filips Ch (2003) Ultraschallsignalverarbeitung bei Korrelations- und Vortexverfahren zur Durchﬂußmessung. Dissertation, Universität Essen, also published by Cuvillier Verlag Göttingen (ISBN 3-89873-711-X) Filips Ch (2001) Tomographische Darstellung des Strömungsproﬁls mit wenigen Daten. Techn. Messen 68:226–233 Gersten K, Herwig H (1992) Strömungsmechanik - Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig, Wiesbaden Kolpatzik SJ (1999) Numerische Simulation der Ausbreitung von Ultraschallwegen in Gasströmungen. Dissertation, Universität Essen, also Fortschritt-Berichte VDI, Reihe 7, 373 Massa (1999) Datasheet E-188. Hingham, MA, USA Niemann M (2002) Signalverarbeitung in der Ultraschall-Durchﬂussmessung. Dissertation, Universität Essen Nikuradse J (1932) Gesetzmäßigkeiten der turbulenten Strömung in glatten Rohren. VDIForschungsheft 356 Papoulis A (1984) Probability, Random Variables, and Stochastic Processes. Mc Graw Hill Poppen G (1997) Durchﬂußmessung auf der Basis kreuzkorrelierter Ultraschallsignale. Dissertation, Universität Essen Rettich T (1999) Korrelative Ultraschall-Durchﬂussmessung auf Basis turbulenter Strukturen. Dissertation, Universität Essen also VDI-Fortschr.-Ber., Reihe 7, Nr. 359, VDI-Verlag Düsseldorf Schneider F (2001) Eine Analyse der Entstehung der Messsignale bei der korrelativen Ultraschall-Durchﬂussmessung in turbulenter Strömung. Dissertation, Universität Essen Schrüfer E (1995a) Elektrische Messtechnik. Hanser-Verlag, München Schrüfer E (1995b) Signalverarbeitung. Hanser-Verlag, München Shu W (1987) Durchﬂussmessung in Rohren mit Hilfe von künstlichen und natürlichen Markierungen. Dissertation, Universität Karlsruhe Skwarek V (2000) Verarbeitung modulierter Ultraschallsignale in Ein- und Mehrpfadanordnungen bei der korrelativen Durchﬂußmessung. Dissertation, Universität Essen Skwarek H, Hans V (1999) Modern principles of signal processing for analysing properties of turbulent ﬂuids. WISP 99, Proceedings:5–10 Skwarek H, Windorfer H, Hans V (2001) Measuring pulsating ﬂow with ultrasound. Measurement 29/3:225–236 Skwarek H, Hans V (2000) Multipath cross-correlation ﬂow meters. IMEKO XVI, Proceedings, TC 9 Worch A (1998) A clamp-on ultrasonic cross-correlation ﬂow meter for one-phase ﬂow. Meas. Sci Techn. 9:622–630

8 Ultrasound Cross-Correlation Flow Meter: Analysis by System Theory and Inﬂuence of Turbulence Franka Schneider, Franz Peters, Wolfgang Merzkirch Institut für Strömungslehre, Universität Essen, 45117 Essen, Germany

The physical principles of the ultrasound cross-correlation ﬂow meter are described quantitatively by means of an approach based on system analysis. We extend an earlier theoretical model of Shu (1987) by accounting for the characteristics of the turbulent pipe ﬂow and making use of characteristic turbulence parameters. The velocity obtained by cross-correlation of the phase-modulated ultrasound signals can be predicted with the theoretical model as a function of the Reynolds number. These theoretical predictions agree favorably with experimental results. It is shown that the velocity determined by cross-correlation is nearly the “modal value” of the turbulent velocity proﬁle and, thereby, higher than the bulk velocity in the pipe. The investigations include studies of the inﬂuence of a number of parameters that can be varied in a practical setup. It is shown that a calibration of the system, as is done in practice, is only valid for a speciﬁc set of values of these parameters and restricted to fully developed ﬂow in the pipe.

8.1 Introduction The principles of ﬂow metering by means of ultrasonic cross-correlation were communicated, e.g., by Mesch (1982) and Beck (1987); see also Chap. 7 of this book. This method of measuring the volumetric ﬂow rate in a pipe is an interesting approach because the method is nonintrusive, does not need information on the speed of sound of the ﬂuid, and it can be applied to single- and multiphase ﬂow. In such a ﬂow meter two ultrasonic waves (or rays), separated by an axial distance L, are transmitted continuously across the pipe, from wall to wall and normal to the pipe axis (Fig. 8.1). The two signals received continuously in time are cross-correlated. The position of the maximum of the cross-correlation function is interpreted as being the time needed by markers in the ﬂow for being convected from the upstream to the downstream ultrasonic “ray”. The distance L divided by this convection time gives a convection velocity that is calibrated in terms of the bulk velocity (volume ﬂow rate divided by the pipe cross-sectional area), u ¯b.

130

8 Ultrasound Cross-Correlation Flow Meter: System Theory

s2(t)

L

V

s1(t)

t

t

t

Fig. 8.1. Schematic view of a cross-correlation ultrasonic ﬂow meter applied to pipe ﬂow with the volumetric ﬂow rate V˙ . The ultrasound signals s1 (t), s2 (t), received continuously with time t at the transducers separated by the axial distance L, are cross-correlated. The time t = τ determining the maximum of the correlation function is the delay for a particular pattern of the signal to appear at the downstream position (2), at a distance L from (1).

In multiphase ﬂow the markers can be particles dispersed in the continuous phase. We restrict the investigations to single phase ﬂow for which it is believed that the markers are vortical or coherent structures convected by the turbulent ﬂow of the ﬂuid. The existence of such vortical structures was veriﬁed by experiments using particle-image velocimetry (PIV) (Westerweel et al., 1996; Schneider and Merzkirch, 2001). When a vortex moves through an ultrasound “ray”, the radial velocity component of the vortex is superimposed to the propagation velocity of the sound waves, thus resulting in a phase modulation of the ultrasound wave. Another possible mechanism of the sound wave’s modulation is the pressure ﬁeld associated to the vortical structure. This would result in an amplitude modulation of the wave. The vortical structures are convected in the ﬂow with a velocity that depends on their radial position in the pipe, i.e., the (local) convection velocity is a function of the velocity proﬁle of the turbulent pipe ﬂow. The velocity determined by cross-correlation is associated with the “modal value” (abscissa of the maximum of the PDF) of the wall-to-wall probability density function of the local convection velocities. The ﬁnal value of the volumetric ﬂow rate found by calibration depends on a number of parameters, the most important being the state of ﬂow, e.g., fully developed ﬂow. When the meter is in practical use, these parameter values must coincide with those present at calibration. First descriptions of the ultrasonic cross-correlation ﬂow metering method were presented two decades ago, e.g., Mesch (1982), Beck (1987). Explanations of the measured results were given, e.g., by Manook (1981), Poppen (1997), Worch (1998), Rettich (1999); all these explanations are based on the idea, that the velocity determined by the cross-correlation is the result of an integration of the convection velocities of the vortical structures along the “ray” path. The discrepancy found between the bulk velocity and the so determined (integrated convection) velocity was

8.2 Experiments

131

explained with certain properties of the vortical structures. Because of the lack of statistical data on the vortical structures, e.g., the distribution of the vortical structures in the turbulent pipe ﬂow, these explanations do not include a quantitative analysis of the measuring principle. The principle of ﬂow metering by ultrasonic cross-correlation can be considered as a technical system with the following main components: • • • •

the pipe ﬂow whose spatially and temporally averaged velocity is to be measured, the turbulent vortical structures serving as markers, the ultrasound waves interacting with the markers in the ﬂow, the cross-correlation process used for determining the convection velocity of the markers.

Several theoretical approaches have been reported for describing the principle of this meter by means of system analysis (Kipphan and Mesch, 1978; Kronmüller, 1980; Braun, 1984; Shu, 1987; Shu and Tebrake, 1989). These investigations refer to the general principle of cross-correlation ﬂow metering, independent of the choice of the markers and sensors. These theoretical models were not used for deriving quantitative results that would simulate measurements with the ultrasonic cross-correlation system based on natural markers. In this chapter we describe the development of a model for the ﬂow metering method by means of system analysis. Thereby we follow the strategy used by Shu (1987), we extend this model, and we make use of input data derived from experiments on turbulent pipe ﬂow in order to arrive at quantitative results. In order to compare the analytically derived predictions with measured data, we also perform experiments with the ultrasonic cross-correlation method in pipe ﬂow at Reynolds numbers in the order of 105 . Measurements with particle-image velocimetry (PIV) performed in the same pipe ﬂow serve for providing data on the vortical structures in the ﬂow; the results of these measurements (Schneider and Merzkirch, 2001) are used as input data for the approach by means of system analysis. In the following we ﬁrst describe the ultrasonic ﬂow metering system. Thereafter, the model based on system analysis is described, and this model is then evaluated quantitatively and compared with the ﬂow metering results.

8.2 Experiments Experiments with the ultrasound cross-correlation setup illustrated in Fig. 8.1 are performed with air ﬂow in the pipeline (inner diameter D = 100 mm) that is mentioned in Chap. 4. The measurement position is at 90 diameters downstream of the inlet section, thus securing fully developed ﬂow. The pipe Reynolds number ¯b /ν (ν = kinematic viscosity of air, u¯b = bulk velocity) is varied ReD = D × u between 0.2 × 105 and 2.4 × 105 . Reference measurements of the volumetric ﬂow rate can be recorded continuously in time with an uncertainty of ±0.3% (see also Chap. 4).

132

8 Ultrasound Cross-Correlation Flow Meter: System Theory a.

U [V] 0,1 0,0 -0,1 -0,2

b.

s2(L,t)

s1(0,t) 40

I(L,t)

50

60

70 t [ms]

20

30 t [ms]

W= 7,75 ms

0,4 0,2 0,0 -0,2

c.

0

10

U [V]

0,1 0,0 -0,1 -0,2

s1(0,t+W) 40

50

s2(L,t) 60

70 t [ms]

Fig. 8.2. (a) Example of ultrasound signals s1 , s2 received by sensors (1) and (2), respectively, during a period of time t of 30 ms; dimension of signal amplitude is electric tension U in Volt. (b) cross-correlation function φ12 formed with the two signals s1 , s2 ; (c) superposition of the two signals s1 , s2 after shifting s1 by τ = 7.75 ms; data from Rettich (1999).

For generating and receiving the ultrasonic waves we use piezo transducers with a resonance frequency of 220 kHz. The signals are processed by undersampling at a frequency of 20 kHz. The axial distance between the two ultrasonic “rays” (see Fig. 8.1) is chosen in the majority of the measurements as L = 100 mm(= 1 D) according to the optimisation studies of Rettich (1999). As an example we show in Fig. 8.2a the two signals s1 and s2 received at the two sensors as function of time t for a period of time of 30 ms. The cross-correlation function φ12 with its maximum at τ = 7.75 ms is seen in Fig. 8.2b, while the two signals, superimposed after a shift of s1 by 7.75 ms are shown in Fig. 8.2c. For performing the cross-correlation we use

8.3 System-Theoretical Model

133

the phase-modulated signals. Details of the signal processing are described in Chap. 5 of this book; see also Skwarek (2000). For the performance as well as results of the cross-correlation process we refer to Schneider (2001).

8.3 System-Theoretical Model The measurements reported by Poppen (1997) and Rettich (1999) had shown that the velocity u¯corr determined from the maximum of the correlation function φ12 deviates systematically from the bulk velocity. For giving an explanation and an estimate of this deviation, a model is needed, that can describe the functional relationship ¯b . For this purpose, between the correlation velocity u ¯ corr and the bulk velocity u we interpret the measurement system as a linear time-invariant system. The model described here is based on earlier work of Shu (1987) and Shu and Tebrake (1989). We modify the model with regard to the assumed velocity proﬁle, the diffusion term, and the inﬂuence of the natural ﬂow markers. 8.3.1 Interpretation of the Measurement System as a Linear Time-Invariant System A linear, time-invariant system is deﬁned by time-invariant and linear transmission characteristics. It is postulated here that the ultrasonic cross-correlation ﬂow meter has such properties. Assumption of a true linear, time-invariant system requires the existence of stationary ﬂow between the two sensors, which is given for fullydeveloped, time-averaged turbulent pipe ﬂow. However, in turbulent ﬂow the velocities are time-dependent and stochastic. By forming the cross-correlation function φ12 of the two signals, s1 and s2 , φ12 (τ ) =

1 T

T

s1 (t)s2 (t − τ )dt

(8.1)

0

(with T being the length of the measurement period) it is evident that a temporal average is provided, from which it follows that the system can be considered as being time-invariant in the average (Shu, 1987). The PIV measurements (Schneider, 2001; Schneider and Merzkirch, 2001) showed that the vortical ﬂow structures, i.e., the markers, are relatively small (e.g., in comparison to the pipe diameter). Hence, they allow the application of the superposition criterion of a linear system. For a linear, time-invariant system the signal s2 (t) is linked to the signal s1 (t) by s2 (t) = s1 (t) ∗ h(t)

(8.2)

with h(t) being the impulse response of the system. h(t) expresses the action of the ﬂow on the markers, that are released to the ﬂowing ﬂuid at a given instant of time t1 and at the deﬁned position (1). If the markers are released along the line of a pipe diameter, they will be convected from their initial positions with different

134

8 Ultrasound Cross-Correlation Flow Meter: System Theory

velocities, due to the given velocity proﬁle and diffusion processes occuring in turbulent pipe ﬂow. As a consequence, the impulse response changes its shape along the distance L separating the two measurement positions (1) and (2), which means that h(t) depends also, to some extent, on the axial position x (see Fig. 8.3). With regard to the measurement position (2) at x = L, the impulse response deﬁnes the instants of time for which the markers, convected along different radial coordinates r = const., arrive at position (2), provided they were released simultaneously to the ﬂow at position (1). addition of markers

pipe

flow

x concentration of markers

t0

t=t1>t20

t=t2>t1

x

Fig. 8.3. Schematic representation of the changing shape of the impulse function along the axial coordinate in the pipe ﬂow.

Before we investigate how h(t) can be determined (Sect. 8.3.2), a few general properties of the impulse response will be discussed, mainly with concern to the use of natural markers (turbulent structures), as we consider them here. The following relationships apply to the linear time-invariant system: φ12 = φ11 ∗ h(t)

(8.3)

φ22 = φ11 ∗ h(t) ∗ h(−t)

(8.4)

with φ11 (t), φ22 (t) being the auto-correlation functions of the signals received by the ﬁrst (1) and second (2) sensor, respectively. Equation (8.3) is of interest regarding the position of the maximum of the crosscorrelation function that is used for determining the convection velocity. If the impulse response is symmetric, the time coordinates where the impulse response and the cross-correlation function have their maxima will coincide. If the impulse response is asymmetric, as it is the case in turbulent pipe ﬂow, the shape of the autocorrelation function φ11 (t) inﬂuences the position of the maximum of the crosscorrelation φ12 (t) (see Shu and Tebrake, 1989, and Fig. 8.4). The maxima positions of impulse response and cross-correlation function do not coincide; the difference between the two positions increases with increasing width of the auto-correlation (Fig. 8.4), i.e., with the bandwidth of the signal received by sensor (1).

8.3 System-Theoretical Model

h(t) *

t t

h max

I11 (t)

=

135

I12 (t)

t

t t

h max

tmax

I

Fig. 8.4. Superposition of auto-correlation φ11 and an asymmetric impulse response h(t) results in a difference between the position for the maxima of h(t), thmax , and cross-correlation φ12 , tφmax .

In our measurement system the markers are not released instantaneously to the ﬂow. Instead, the markers (turbulent structures) are convected continuously with the turbulent pipe ﬂow, from which one derives that φ11 (t) = φ22 (t). This is in contradiction to (8.4) or, h(t) ∗ h(−t) = δ(t) (δ = Dirac’s function), i.e., the postulated linear model could be applied only to a measuring system with inﬁnitesimally short distance between the two sensors. In order to account for this problem Shu (1987) introduces into (8.3) a noise term that is linked with φ11 (t) ∗ h(t) in a nonlinear way. This noise term also includes the effect of the continuous formation and disappearance of the turbulent stuctures. There are no means for giving an estimate of the magnitude of the noise term, and only from the quantitative results presented in this paper we may conclude that this term that is neglected in our analysis must be very small. An alternative theoretical model for the considered cross-correlation ﬂow metering systems was reported by Braun (1984). In this model a co-variance matrix is derived that includes a probability density function for describing the transport of the markers. This way, the problem of (8.4) being inconsistent with the real measuring system is not present. However, for determining the maximum of the co-variance matrix function Braun introduces a number of analytical simpliﬁcations regarding certain properties of turbulence; e.g., the probability density function (PDF) of the velocities, which in turbulent pipe ﬂow is skewed, is assumed to have a Gaussian distribution, and diffusion as caused by the turbulent ﬂuctuations is totally neglected. We decided to follow Shu’s approach because it allows us to account for the skewness of the PDF of the convection velocities and to include the turbulent diffusion processes, as it is outlined in the following sections. 8.3.2 Impulse Response of the Ultrasound System The impulse response h(t) needed, according to (8.1), (8.2), for determining the cross-correlation function φ12 (t) and its maximum position depends on a number of parameters characterizing the ﬂow and the setup for taking measurements, e.g., velocity proﬁle of pipe ﬂow, turbulence characteristics, properties of the coherent

136

8 Ultrasound Cross-Correlation Flow Meter: System Theory

structures, sensor size, distance between sensors. The inﬂuence of these parameters on h(t) is studied quantitatively in this section. Convection Process According to their radial position r the turbulent structures (markers) are convected at different (time-averaged) velocities u ¯(r), where u ¯(r) is the velocity proﬁle of the pipe ﬂow for a given Reynolds number. We deﬁne a function Cconvection (x, t) that describes the spatial and temporal distribution of the markers as caused by the convection process. Figure 8.5 illustrates the technical meaning of Cconvection (x, t) under the assumption that the markers are evenly distributed at x = x1 and t = t0 . Cconvection (x, t) gives information on the arrival times of the markers at the position of the second sensor, x = x2 = x1 + L.

Cconvection(x=x1, t=t0) r Cconvection(x, t=t1>t0) Cconvection(x, t=t2>t1>t0)

Cconvection(x=x1+L, t) x Fig. 8.5. Distribution of the markers, Cconvection (x, t), for three different instants of time, t0 , t1 , t2 . The markers are evenly distributed along a pipe diameter at position x = x1 and time t = t0 . The markers are convected in axial direction according to the pipe ﬂow velocity proﬁle, u ¯(r).

According to the transformation laws for probability functions (Schrüfer, 1990), the function Cconvection (x, t) for the convection process is related to the probability density function (PDF) of the velocity, f (¯ u(r)), by . f (¯ u) .. (8.5) Cconvection (x, t) = . dx . . . .. x d¯ u u ¯= t

with u¯ = x/t and 0 ≤ x ≤ u¯max t (¯ umax is the maximum velocity of u¯(r)).

8.3 System-Theoretical Model

137

If one assumes that all structures are detected by sensor (2) with the same local sensitivity and that sensor (2) has an inﬁnitesimally small extension in x-direction, so that its sensitivity can be described with the properties of Dirac’s function, δ(x − L), then the impulse response at sensor (2) can be expressed by ∞ Cconvection (x, t)δ(x − L) dx = kCconvection (L, t)

h(t) = k

(8.6)

0

where k is a constant for normalization, such that

(∞ L/¯ umax

h(t) dt = 1.

The time for which h(t) has a maximum, thmax , deﬁnes a maximum of similarity between the patterns of a turbulent structure convected from sensor (1) to sensor (2). For calculating the PDF f (¯ u(r)) we need an analytical form of the (time averaged) velocity proﬁle u ¯(r). The most complete theoretical description of this proﬁle is the asymptotic (Re→ ∞) theory of Gersten and Herwig (1992). Since their formulation of the proﬁle includes several procedures that cannot be expressed analytically, we use, as a simple alternative, the power law for turbulent pipe ﬂow, and we combine it with a parabolic velocity distribution for the core region according to Hinze (1975). The power law is r n1 u¯(r) = 1− (8.7) u ¯max R with R being the pipe radius and u ¯ max the velocity on the pipe axis. The index n is determined as a function of the Reynolds number (Nikuradse, 1932): 1 = 0.25 − 0.023 log(Re) n

(8.8)

in the range 4 × 103 ≤ Re ≤ 3.2 × 106 . The parabolic velocity distribution according to Hinze makes use of the friction velocity uτ : r 2 u ¯max − u¯ = 7.2 (8.9) uτ R The two proﬁles are combined without adjusting their slopes, so that a discontinuity results when forming the ﬁrst derivative. In order to determine the PDF of u ¯(r) according to Schrüfer (1990) . . . . 1 . , (8.10) |f (¯ u(r))| = ..f (r) d¯ u(r)/dr . we must know the probability distribution of the turbulent structures f (r). f (r) depends on the distribution of the structures across the pipe cross section and the area range for which the sensor collects the signal information. If the structures are distributed uniformly in a cross section, and the sensor is sensitive with

138

8 Ultrasound Cross-Correlation Flow Meter: System Theory

respect to the whole cross section (“area sensor”), one has f (r) = 2r/R 2 . For a sensor collecting the information on the structures along a pipe diameter (“line sensor”), one has f (r) = 1/R. The ultrasound transducers used in our experiments have a diameter of 10 mm. We assume the detection volume of these sensors to be a 10 mm wideband through the pipe axis. In determining f (r) we take care of the distribution of the markers in this detection volume. Equations (8.10), (8.7) resp. (8.9), and (8.5) allow to determine the impulse response (8.6) in accounting for the pure convection process. Diffusion Process The ﬂuctuating motion of turbulent ﬂow causes the turbulent structures to diffuse along their path from sensor (1) to sensor (2). In studying the inﬂuence of this diffusion process on the signal formation, a few simplifying assumptions will be made. Diffusion is considered in streamwise direction only, which appears to be justiﬁed if the distance L between the sensors is small enough such that, for their convection along L, the structures keep their initial radial position r. Furthermore, we assume as an approximation, that the probability density function (PDF) of the velocities u(r, t) is Gaussian (which we did not assume concerning the convection of the turbulent structures): 2 (u(r, t) − u ¯(r)) 1 exp − . (8.11) f (u(r, t)) = √ 2σu 2 2πσu σu is the standard deviation of the velocity u(r, t) in x-direction, and with separating u(r, t) into a time-averaged (mean) value u¯(r) and a ﬂuctuation u (r, t), u(r, t) = '

u 2 (RMS). For turbulent pipe ﬂow σu is known to u ¯(r) + u (r, t), one has σu = be a function of r and Reynolds number; 'see, e.g., Laufer (1954), Perry and Abell (1975). An exact analytical function for u 2 is not available, and therefore we use a quadratic approximation that is supported by PIV measurements performed in our laboratory and by data from the literature (Nikuradse, 1932; Laufer, 1954): ' r 2 r u 2 + 0.85 (8.12) = + 0.5 uτ R R ' where the shear velocity uτ = τw /ρ includes the wall shear stress that is Reynolds number dependent. According to the transformation laws for probability functions (Schrüfer, 1990) the inﬂuence of turbulent diffusion on the distribution of the structures is then described by (cf. (8.5)) . 2 1 (x − u ¯ t) f (u) .. =√ exp − . (8.13) Cdiﬀusion (x, t) = . dx . . . .. 2σu 2 t2 2πσu t x du u= t

In contrast to Shu (1987) who takes σu as a constant we account for the dependence of σu on r according to (8.12).

8.3 System-Theoretical Model

z

diffusion

z

flow

139

x x y

y

velocity profile

Fig. 8.6. Superposition of axial diffusion to the convection process.

Superposition of Convection and Diffusion The distribution considering diffusion of the structural patterns as given by (8.13) must be combined with the results obtained for the convection process, i.e., the PDF of the (time-averaged) velocity u ¯(r) according to (8.10); (see Fig. 8.6). As a result one obtains for the distribution of the structures u ¯max

C(x, t) =

f (¯ u(r))Cdiﬀusion (x, t) d¯ u.

(8.14)

0

The integral must be solved numerically. Figure 8.7 shows the pattern of the impulse response function h(t) when C(x, t) according to (8.14) is used for calculating h(t); the assumed parameter values are given in the ﬁgure legend. For comparison, h(t) is also shown if only the convection processes are considered. It is seen that the proﬁle

0,18 0,16

convection

0,14

h(t) [1/s]

0,12 0,10 0,08 0,06

convection & diffusion

0,04 0,02 0,00 0,005

0,006

L/umax

0,007

tL=L/ub

0,008

0,009

0,01

0,011

t [s]

Fig. 8.7. Impulse response h(t) calculated in accounting for pure convection and for convection plus diffusion. This example was calculated with the assumed parameter values L = D = 100 mm, Re = 105 , and it serves only for demonstrating the characteristic pattern of h(t).

140

8 Ultrasound Cross-Correlation Flow Meter: System Theory

of h(t) is widened towards smaller values of t when diffusion is taken into account, and that its maximum is then shifted towards larger values of time t. t L in Fig. 8.7 is the time that would result if the measuring system would indicate the bulk velocity ¯ b. u ¯b . It is seen that the velocity derived from the maxima positions is higher than u

uhmax /ub

1.32 1.30

h(t), convection, L=1D

1.28

h(t), convection & diffusion, L=1D

1.26

ccf, convection & diffusion, L=1D

1.24 1.22

ucorr /ub

1.20 1.18 1.16 1.14 1.12 1.10 1.08 1.06 0

50000

100000

150000

200000

250000

Re

Fig. 8.8. Non-dimensional velocity u ¯hmax /¯ ub calculated by using the maximum position of ub calculated by using the the impulse response h(t), and non-dimensional velocity u ¯corr /¯ maximum position of the cross-correlation function calculated according to (8.3). Assumend distance between sensors L = D = 100 mm (ccf=cross-correlation function).

The effect of taking the diffusion process into account is also demonstrated in Fig. 8.8 where the velocity u¯hmax = L/thmax , determined by using the maximum position of the impulse response h(t), is shown as function of the Reynolds number, again calculated with the assumed values L = D = 100 mm. u¯ hmax is made nondimensional by means of the bulk velocity u¯b . u¯hmax is shown for the two cases that only convection and convection with superimposed diffusion are considered. The inclusion of diffusion shifts the calculated velocity towards lower values, as was visible in Fig. 8.7. The asymmetry of the impulse response (cf. Fig. 8.4) causes the calculated velocity values to be too high in comparison to the velocity derived from the maximum position of the cross-correlation function. This is evident from the lower curve in Fig. 8.8 that was determined by making use of φ12 (t) according to (8.3). Since theoretical values of the auto-correlation φ11 (t) are not available, it was necessary to use empirical data of φ11 (t) from our measurements. For each value of Reynolds number we used an auto-correlation function derived from a total of 50 individual measurements. Inﬂuence of the Distribution of the Natural Markers In the previous sections we have assumed, that the markers are uniformly distributed in a pipe cross section. The PIV measurements of Schneider and Merzkirch (2001)

8.4 Comparison of Theoretical and Experimental Results

141

have shown, that this assumption is, in principle, justiﬁed. As a difference to artiﬁcial markers, natural markers might have properties, that depend on their position in the pipe, e.g., the turbulence intensity and the life time of the coherent structures. These properties can have an inﬂuence on the signal generation, if they interact with the ultrasound wave that depends on the selected sensor type. Interaction can occur between the ultrasound and turbulent ﬂuctuations (Chernov, 1960; Tatarski, 1971), and of particular interest here is the ﬂuctuation component in the direction of the ultrasound propagation, v . We account for this effect by weighting the distribution of the markers at the 'position of the ﬁrst sensor, according to the radial distribution of the RMS values v 2 /uτ as given by Eggels et al. (1994). This has a small, but visible effect on the shape of the impulse response as reported in more detail in Schneider (2001).

8.4 Comparison of Theoretical and Experimental Results In this section we compare predictions calculated with the model presented in Sect. 8.3 with experimental results measured in the facility mentioned in Sect. 8.2. The aim is to validate the model, explain the observed difference between measured velocity and bulk velocity, and study the inﬂuence of a number of experimental parameters on the signal. A set of measured cross-correlation functions and the cross-correlation calculated with the model is shown in Fig. 8.9 for the experimental conditions speciﬁed in the ﬁgure legend. The cross-correlation function is shown here in the form (“cross-correlation coefﬁcient”) φ12 (t) =

φ12 (t) . φ11 (0)φ22 (0)

(8.15)

The asymmetry that was explained in Sect. 8.3.1 and Fig. 8.3 is visible, and it can also be seen that the positions of the maxima of the measured functions are, to a small extent, at higher values of time than that for the theoretical curve, which is in agreement with the results shown in Fig. 8.10. The non-dimensional velocity determined by using the maximum position of the ub , is shown in Fig. 8.10 as function of the (pipe) cross-correlation function, u ¯ corr /¯ Reynolds number. Here, the results of the theoretical predictions are presented together with the results of the measurements (cf. Sect. 8.2). The vertical bars indicate the scatter of the experimental data for a set of approximately 10 measurements for each Reynolds number. For comparing the dependence on Reynolds number we also ub calculated with show in this ﬁgure the non-dimensional maximum velocity u ¯ max /¯ Gersten and Herwig’s (1992) theory. The difference between predicted and measured velocities is in the order of 3 to 4%, and this can be taken as proof of the validity of the theoretical model. In the previous sections we mentioned a number of phenomena of minor importance that are not included in the theoretical model and possibly the reason for the 3 to 4% deviation. The position of the maximum of the crosscorrelation function, tφmax depends on the position of the maximum of the impulse

142

8 Ultrasound Cross-Correlation Flow Meter: System Theory 0.8

experiments model

0.7 0.6

I12

0.5 0.4 0.3 0.2 0.1 -0.1 -0.2 0.000

0.002

0.004

0.006

0.008

0.010

0.012

t [s]

Fig. 8.9. Non-dimensional cross-correlation function φ12 (t): - - - - experiments, —— theoretical model. Parameter valus: Re = 105 , L = D = 100 mm.

ucorr/ ub , umax/ ub

response function, thmax , as shown in Fig. 8.4. thmax is the time at which a maximum of markers arrive at sensor (2) after they had been released at sensor (1) and time t = 0, i.e., L/thmax indicates a velocity whose presence in the velocity proﬁle has the highest probability (“modal value”), expressed by (8.10). Therefore, one has always u ¯corr /¯ ub > 1 as seen in Fig. 8.10. 1.28 1.26 1.24 1.22 1.20 1.18 1.16 1.14 1.12 1.10 1.08 1.06 1.04 1.02 1.00

experiment model umax/ubb umax/u

0

50000

100000

150000

Re

200000

250000

Fig. 8.10. Theoretical and experimental results of the velocity, found by cross-correlation, as function of Reynolds number. Values of the maximum velocity are given for comparison. Distance between sensors L = D = 100 mm.

A parameter that can be varied in the experimental setup is the axial distance between the two sensors, L (cf. Fig. 8.1). Measured and calculated values of the ub , are presented in non-dimensional velocity derived by cross-correlation, u¯ corr /¯

8.4 Comparison of Theoretical and Experimental Results 1.22

143

experiment, L = 2D experiment, L = 1D experiment, L = 0.6D model, L = 2D model, L = 1D model, L = 0.6D

1.20 1.18 1.16

ucorr/ ub

1.14 1.12 1.10 1.08 1.06 1.04 1.02 1.00 0

50000

100000

150000

200000

250000

Re

Fig. 8.11. Inﬂuence of the distance between sensors, L, on the velocity determined by crosscorrelation (experiment and theoretical model) for different Reynolds numbers.

Fig. 8.11 as function of the Reynolds number. The measured curves exhibit certain irregular oscillations, that might be caused by instabilities in the electronic signal processing system. Not much attention should be given to these irregularities at this ub on the distime. The interesting result in Fig. 8.11 is the dependence of u¯ corr /¯ tance L, both in the experiment and the theoretical prediction. This indicates that a calibration of a setup for taking measurements of the volumetric ﬂow rate is only ub on L can valid for a speciﬁed sensor distance L. The shown dependence of u ¯ corr /¯ be explained with the changing distribution of the markers when they are convected in axial direction (cf. Fig. 8.5), which makes the impulse response function in the theoretical model to depend on the axial position x, thereby affecting the maximum position of the cross-correlation function. The absolute value of the maximum of the cross-correlation is an indication of the degree of correlation. The maximum value of the cross-correlation coefﬁcient, both for experiment and theory, is presented in Fig. 8.12 as function of the Reynolds number and for three values of the distance L between the sensors. The results shown conﬁrm the high degree of agreement between experiment and theoretical model. The slight increase of the maxima with Reynolds number can be explained with the velocity proﬁle that becomes more ﬂat with increasing Reynolds number; this causes more markers to be convected with almost the same velocity, thus increasing the degree of similarity of the received signals. The similarity of the signals decreases with increasing L due to the changing distribution of the markers in axial direction, as it was mentioned before and illustrated in Fig. 8.5. Figure 8.12 also indicates that, for the range of values of L varied in these experiments, the highest degree of correlation is reached here for L = 0.6. If one decreases L further below 0.6, the degree of correlation will decrease again due to an undesired mutual interaction

144

8 Ultrasound Cross-Correlation Flow Meter: System Theory 0.9 0.8 0.7

I12max

0.6 0.5 0.4

experiment, L = 2D experiment, L = 1D experiment, L = 0.6D model, L = 2D model, L = 1D model, L = 0.6D

0.3 0.2 0.1 0.0 0

50000

100000

Re

150000

200000

250000

Fig. 8.12. Absolute values of the maxima of the cross-correlation coefﬁcient as function of Reynolds number for three different distances between sensors, L.

(cross-talk) of the ultrasound transducers at positions (1) and (2). We did not attempt to determine the optimum value of L regarding a maximum of correlation. Experiments were also performed with a setup in which the receiver at the second sensor position was not perpendicularly opposite to the ultrasound emitter. The used geometry is shown in Fig. 8.13. These experiments served only for exploring the capabilities of the theoretical model. The calculated cross-correlation coefﬁcient φ12 (t) shown in Fig. 8.14 for a speciﬁc set of experimental parameters (Re = 10 5 , D = 100 mm) has two maxima, from which one can conclude that this setup is not useful for practical measurements of the volumetric ﬂow rate. The occurance of the two maxima can be explained with the theoretical model. We have derived that the velocity determined by cross-correlation is nearly the “modal value” of the velocity proﬁle u ¯(r). In Fig. 8.13 it is seen that the ultrasound “ray” of the oblique sensor (2) interacts twice with the “modal value”, but for different axial distances between the sensor “rays” (1) and (2), thus causing the occurance of two maxima in the cross-correlation function. With (8.11) one derives that, for Re = 10 5 , the most

S1

0.6D

S2

0.25D

1D

0.25D

E1

1D

E2

Fig. 8.13. Setup with oblique ultrasound “ray” for sensor (2).

8.4 Comparison of Theoretical and Experimental Results

145

0.5 0.4 0.3

I12

0.2 0.1 0.0 -0.1 -0.2 0.000

0.002

0.004

0.006

0.008

0.010

t [s]

Fig. 8.14. Cross-correlation coefﬁcient obtaind with the setup shown in Fig. 8.13 for Re = 105 , L(S1 − S2 ) = 0.6D, D = 100 mm.

frequently occuring velocity (“modal value”) is in the range from u ¯ corr ≈ 1.06¯ ub to ¯(r) this value is found at the position 1.08¯ ub, and in the turbulent velocity proﬁle u r ≈ 0.25D. Using these data and the geometry shown in Fig. 8.13 one veriﬁes the temporal positions of the two maxima of the cross-correlation coefﬁcient shown in Fig. 8.14. In the beginning it was mentioned that the interaction of the ultrasound wave with the turbulent ﬂow can also result in an amplitude modulation of the wave. The form of the cross-correlation function depends on whether the phase modulation or the amplitude modulation is used for the signal generation (Manook, 1981; Skwarek, 1.2

acf of the phase signal 1.0

acf of the amplitude signal

I11

0.8 0.6 0.4 0.2 0.0 -0.2 -0.005 -0.004 -0.003 -0.002 -0.001 0.000

0.001

0.002

0.003

0.004

0.005

t [s]

Fig. 8.15. Auto-correlation coefﬁcient derived from phase-modulated and amplitudemodulated signals (Re = 105 ) (acf=auto-correlation function).

146

8 Ultrasound Cross-Correlation Flow Meter: System Theory 1.19

phase modulated signal

1.18

amplitude modulated signal

1.17

ucorr/ ub

1.16 1.15 1.14 1.13 1.12 1.11 1.10 1.09 0

50000

100000

150000

200000

250000

Re

Fig. 8.16. Theoretical results of the non-dimensional velocity determined by cross-correlation by using the phase-modulated and amplitude-modulated signals (L = D = 100 mm).

2000). The auto correlation φ11 (t) obtained from the amplitude-modulated signals is narrower than φ11 (t) derived from phase-modulated signals as shown in Fig. 8.15 for a special set of parameters (Re = 105 ). From Fig. 8.4 it follows that the position of the maximum of the cross-correlation function is then closer to the maximum position of the impulse reponse h(t), i.e., at lower values of time, and the resulting velocity must be higher than the values determined by using the phase-modulated signals. This is in agreement with the experimental ﬁndings of Skwarek (2000) who ub are about 3% higher than those found by reports, that the velocity values u ¯ corr /¯ using the phase-modulated signals. This observation is validated quantitatively with our theoretical model, as it can be seen in Fig. 8.16. A reason for not using the amplitude-modulated signal is that these signals are more distorted than those obtained from phase modulation (see Fig. 8.17, and Schneider, 2001; Skwarek, 2000). 1.30 1.25

ukorr/ub

1.20 1.15 1.10 1.05 1.00

phase modulated amplitude modulated

0.95 0.90 5

10

15

20

25

30

35

ub [m/s]

Fig. 8.17. Non-dimensional velocity determined by cross-correlation in turbulent pipe ﬂow using either the phase-modulated or the amplitude-modulated ultrasound signals; u ¯b = bulk velocity from the reference measurements; data from Skwarek (2000).

References

147

8.5 Conclusions The theoretical model based on system analysis gives a complete physical description of the ultrasound cross-correlation method for determining the volumetric ﬂow rate in a pipe. We have extended the model whose principles were described by Shu (1987) and Shu and Tebrake (1989) by introducing additional terms characterizing the ﬂow and by making use of characteristic data of turbulent pipe ﬂow. With the theoretical model it was possible to derive quantitative data that could be compared with experimental results. The theoretically and experimentally determined results for the velocity resulting from cross-correlation of the phase-modulated ultrasound signals deviate by 3 to 4%. The difference must be attributed to the approximation made in (8.2) regarding an additional noise term and to phenomena that, at this time, are difﬁcult to account for in the theoretical model. Because of this difference it is impossible to present here a theoretical estimate of the measurement accuracy of the method. From the theoretical model it follows that the velocity obtained by crosscorrelation is nearly the “modal value” of the time-averaged turbulent velocity proﬁle in the pipe. With (8.10) this “modal value” can be determined if the velocity proﬁle is known, as is the case for fully developed ﬂow. The impulse response depends on the distance of the sensors L, see (8.5) and (8.6), i.e., a calibration of the measurement system is only valid for the value of L used at calibration. Any change of L in a measurement setup would require a new calibration. Also, the calibration only applies to the state of ﬂow present during the calibration process. A system using a multiple-path arrangement is probably not an improvement of the situation, since the signal received is not the result of an integration of the velocity values from wall to wall. With the theoretical model it was also possible to explain a number of phenomena that have been observed in experiments with setups that turned out to be not appropriate for quantitative measurements of the volumetric ﬂow rate. Finally, an answer could be given to the question whether it is more useful to derive the crosscorrelation results from the phase-modulated or the amplitude-modulated ultrasound signals. The answer is clearly that the phase-modulated signals are the better choice.

References Beck MS (1987) Cross correlation ﬂowmeters: their design and application. IOP Publishers, Bristol Braun H (1984) Statistik der Signale berührungsloser Strömungssensoren. Dissertation, Universität Karlsruhe (TH) Chernov L (1960) Wave Propagation in a Random Medium. McGraw-Hill, New York Eggels JG, Unger F, Weiss MH, Westerweel J, Adrian RJ, Friedrich R, Nieuwstadt FTM (1994) Fully developed turbulent pipe ﬂow: a comparison between direct numerical simulation and experiment. J Fluid Mech 268:175–209 Gersten K, Herwig H (1992) Strömungsmechanik - Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig

148

8 Ultrasound Cross-Correlation Flow Meter: System Theory

Hinze JO (1975) Turbulence, 2nd edition. McGraw-Hill, New York Kipphan H, Mesch F (1978) Flow measurement systems using transit time correlation. In: Flow Measurement of Fluids, eds. Dijstelbergen HH, Spencer EA, 409–415, North-Holland Publishing Company Kronmüller H (1980) Durchﬂußmessung mittels Markierungsverfahren. VDI-Berichte Nr. 375:47–54 Laufer J (1954) The structure of turbulence in fully developed pipe ﬂow. NACA Report 1174 Manook BA (1981) A high resolution cross-correlator for industrial ﬂow measurement. Dissertation, University of Bradford Mesch F (1982) Geschwindigkeits- und Durchﬂussmessung mit Korrelationsverfahren. Regelungstechn Praxis 24:73–82 Nikuradse J (1932) Gesetzmäßigkeiten der turbulenten Strömung in glatten Rohren. VDIForschungsheft 356 Perry AE, Abell CJ (1975) Scaling laws for pipe-ﬂow turbulence. J Fluid Mech 67:257–271 Poppen G (1997) Durchﬂußmessung auf Basis kreuzkorrelierter Ultraschallsignale. Dissertation, Universität Essen Rettich T (1999) Korrelative Ultraschall-Durchﬂussmessung auf Basis turbulenter Strukturen. Dissertation, Universität Essen (also published as VDI-Fortschr.-Ber., Reihe 7, Nr. 359, VDI-Verlag Düsseldorf) Schneider F (2001) Eine Analyse der Entstehung der Messsignale bei der korrelativen Ultraschall-Durchﬂussmessung in turbulenter Strömung. Dissertation, Universität Essen Schneider F, Merzkirch W (2001) Distribution of coherent structures in turbulent pipe ﬂow. J Flow Visual Image Processing 8:253–261 Schrüfer E (1990) Signalverarbeitung. Hanser Verlag, München Shu W (1987) Durchﬂussmessung in Rohren mit Hilfe von künstlichen und natürlichen Markierungen. Dissertation, Universität Karlsruhe Shu W, Tebrake G (1989) Durchﬂußmessung in Rohren mit Hilfe von künstlichen und natürlichen Markierungen. Techn Messen 56:58–65 Skwarek V (2000) Verarbeitung modulierter Ultraschallsignale in Ein- und Mehrpfadanordnungen bei der korrelativen Durchﬂußmessung. Dissertation, Universität Essen Tatarski VI (1971) The effects of turbulent atmosphere on wave propagation. IPST Keter Press, Jerusalem Westerweel J, Draad AA, van der Hoeven JGT, van Oord J (1996) Measurement of fully developed turbulent pipe ﬂow with digital particle image velocimetry. Exp Fluids 20:165– 177 Worch A (1998) A clamp-on ultrasonic cross correlation ﬂow meter for one-phase ﬂow. Meas Sci Technol 9:622–630

9 Effect of Area Changes in Swirling Flow Venkatesa Vasanta Ram Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

This chapter addresses the question of the effect of changes in the ﬂow-passage area on the radial distribution of the velocity vector, in other words on the proﬁles of the axial and the azimuthal velocity components, in axisymmetric incompressible swirling ﬂow. The problem is approached by examining the dynamics of the ﬂow through Euler’s equations of motion. The method of examination makes explicit use of a well-established physical property of invariants in this ﬂow. This is that the total pressure and the moment of azimuthal velocity remain constant along a (cylindrical) stream surface. The study illuminates the nature of the change in the proﬁles of the axial and azimuthal velocity distribution, which involve changes of both scale and shape. Notation x, r, φ co-ordinate system: axial, radial and azimuthal directions respectively ρ density of the ﬂowing ﬂuid ux , ur , uφ velocity components in x, r and φ directions respectively H total head, (9.5) C moment of azimuthal velocity, (9.6) R(x) radius of ﬂow passage of circular cross section R1 , R2 radius of ﬂow passage at inlet and outlet sections Q volume ﬂow rate through the passage πR2 Uref bulk velocity, reference velocity, deﬁned through Q = Uref Ω angular velocity ψ ψ = constant deﬁnes stream surface Ψ value of ψ at outer boundary deﬁned in (9.26) ψ ∗ , r∗ parameter characterizing inhomogeneity, (9.28), (9.29) in

150

9 Effect of Area Changes in Swirling Flow

9.1 Introduction Many engineering applications involve metering of the ﬂow in a pipe. A problem often faced in this task is that, for some reason or other, there often exists some swirl in the ﬂow to be metered, and swirl introduces an error in metering. The swirl present is often traceable to factors of installation that generate a swirling motion of the ﬂuid, which is a kind of motion to be deﬁned shortly. Some common examples of such installational elements are: an elbow or a succession of bends in the pipe, an axial compressor installed upstream for boosting purposes, and so on. Swirling motion, once it arises, has the property of decaying only rather slowly, and installational constraints do not often permit a sufﬁcient length of piping for the ﬂow to settle down sufﬁciently to a state mandatory for correct metering. While it is true of any metering device that it can only “read” the ﬂow it is “exposed” to, there are some peculiarities in the operation of a turbine ﬂow meter that call for a closer examination of the effect of area changes on the ﬂow characteristics. The peculiarities just referred to are derivable from the physical principles underlying conversion of the ﬂow rate into a signal in the turbine ﬂow meter, and these are the subject of Chap. 10. Here, it sufﬁces to note that, after the swirling motion is originated, the ﬂow will have undergone some modiﬁcation on passing through sections of changing effective area prior to being “perceived” by the ﬂow meter. For this reason the effects of area changes on the ﬂow warrants closer study, and this may be identiﬁed as the ﬁrst of two distinct steps in the engineer’s task of assessing the response of the ﬂow meter. Its signiﬁcance for a correct estimation of the error in metering caused by swirl in the ﬂow cannot be overemphasised, neither can it be overlooked that this step involves a deeper understanding of ﬂuid ﬂow phenomena. The relevance to practical applications is better appreciated on noting that in many applications there is a relatively short section of a pipe introduced between the swirl generating element of the installation, such as the elbow or the booster, and the ﬂow meter. This section of the pipe, while being insufﬁcient to condition the ﬂow to the required standards for metering, has boundary layers growing on its walls that constrict the effective area available for the ﬂow passage. The second and subsequent step of answering the question of the signal response of the ﬂow meter to the ﬂow it “perceives”, which is at the exit section of this pipe, is conceptually different, hence meaningful to be treated separately. Here in the present chapter we address the question in the ﬁrst step, viz. of the ﬂow response to area changes. Flows with swirl have a long tradition in ﬂuid mechanics research, in the course of which many of the diverse physical phenomena herein have been uncovered, addressed and explained. Section 9.2 titled Physical Background attempts to provide a proper perspective for the present work against this background. The kinematic and dynamic features of a ﬂow with swirl that are relevant to metering are summarised for preparatory purposes, and the scope of the present work deﬁned. Section 9.3 that follows carrying the title Mathematical Formulation of the Problem is the main part of the work. It sets out the equations and boundary conditions for the problem, introduces the approximation on which the rest of the work is based and outlines the

9.2 Physical Background

151

method of solution. The subsequent section contains sample results that illustrate the applicability of the method in the context of ﬂow metering.

9.2 Physical Background The term “swirl” has been used here to denote a certain outstanding kinematic feature of the ﬂow. It is that in a polar coordinate system placed with its axis coinciding with the pipe axis which is the natural choice for a pipe ﬂow, the azimuthal component of the velocity vector is not zero. The salient characteristic feature of a ﬂow with swirl is therefore deﬁned through the topology of its streamlines. The non-zero azimuthal velocity component in conjunction with the axial velocity component that is decisive for the ﬂow rate results in helically wound streamlines. Dynamically, swirl in the ﬂow is associated with a radial pressure gradient of a strength that depends upon the square of the magnitude of the azimuthal velocity component and the curvature of the projection of the streamlines on to a plane perpendicular to the pipe axis. 9.2.1 Kinematic and Dynamic Characteristics Swirling motion in a pipe may occur in widely differing patterns. For an initial classiﬁcation of the same one may adopt a kinematical standpoint according to which a suitable criteron would be the change of sign, if any, of the (non-zero) azimuthal velocity component along the circumference. No change of sign of this quantity is tantamount to the region of swirling motion extending over the entire cross section with one center identiﬁable. Alternatively, the swirling motion can also occur in cells in which case more than one center is discernible around which the ﬂuid particles describe a “swirling motion”. This is a kind of motion in which the ﬂuid particles apparently describe closed circuits; apparent, since it appears so only in the projection and the streamlines are not in fact closed on themselves. It is however important to note that, regardless of the means of swirl generation, the ﬂow is essentially threedimensional, and the shape of the streamlines is closer to helices than to straight lines. The actual pattern of the swirling motion assumed by the ﬂow in a pipe depends on various factors, with geometrical features and ﬂow parameters both contributing likewise to its formation. Helical streamlines characteristic of swirling motion also imply that the streamline curvature is spatial as opposed to planar. Given the numerous possibilities of swirl generation, it is obvious that a wide variety of ﬂow patterns is permissible with which the ﬂow can enter the pipe section upstream of the ﬂow meter. The attendant characteristics of the velocity and pressure distributions are also therefore multifarious. The velocity distribution established at the inlet section of this pipe then evolves to the state at its end that, in turn, is perceived by the ﬂow meter at its inlet. The evolution of the ﬂow in the pipe from one section to the other is governed by the physical laws of ﬂuid motion, and these are required to be properly taken into account in analyzing the effect of swirl on metering.

152

9 Effect of Area Changes in Swirling Flow

From the standpoint of the dynamics of the ﬂow, the swirling ﬂow in a pipe may be considered as a special case of a more general class of ﬂows characterized by their topological feature of helically wound streamlines. Flows with helically wound streamlines are of inherent scientiﬁc interest due to the phenomena they exhibit, of which vortex breakdown holds a key position. They are also of relevance to understanding ﬂow phenomena both in nature and in a wide variety of engineering applications. It is therefore only very natural to ﬁnd ﬂows with helical streamlines being the subject of extensive studies in ﬂuid mechanics. Literature on this subject is accordingly voluminuous, and we conﬁne ourselves here to citing here only a short selection from recent publications that are representative of the questions addressed, Brown and Lopez (1990), Buntine and Saffman (1995), Escudier (1988), Faler and Leibovich (1978), Gallaire and Chomaz (2003), Rusak et al. (1998), Wang and Rusak (1997). Fundamental properties of these ﬂows that are of signiﬁcance for the objective pursued here have been established in these studies, and these provide the foundation and a convenient starting point for our analysis. Studies hitherto, see eg. Chap. 4 in this book, have focussed attention on to several features exhibited by ﬂows with swirl in a pipe that are of importance in the context of ﬂow metering. Perhaps the ﬁrst and foremost among these, is that swirl, when it arises, decays only rather slowly, see also Chap. 2. It may be noted in this context that swirl in a ﬂow, when understood to be characterized through helical winding of streamlines, does not necessarily imply rotational symmetry. The reader is referred to Chaps. 2, 14 and 15 in this book for examples of rotational asymmetry arising in the ﬂow. However, the special case of the rotationally symmetric ﬂow with swirl is of special signiﬁcance for various reasons. Firstly, as already stated earlier, the axisymmetric mode is the one with the lowest rate of decay, hence likely to persist longest in the ﬂow. Secondly, being simpler, the axisymmetric case permits analysis to be taken further than the others, thus enabling deeper physical insight to be gained into the mechanisms at work, of course in axisymmetric ﬂow. Strictly speaking, the effect of the area change on proﬁles of both the mean and the ﬂuctuating velocity components should be sought since the metering device, depending upon the physical principle on which it is based, could be sensitive to one or both. But in the present work we restrict our attention to the mean proﬁles alone. Again, even with this restriction, we conﬁne ourselves to the inviscid ﬂow. It is appropriate to note at this juncture that a complete picture of the effects of area changes on the ﬂow can of course be obtained only after supplementing the insight gained from the study of the inviscid ﬂow through a study of the boundary layer. Since knowledge of the inviscid ﬂow is a prerequisite for studying the boundary layer, the present study should be regarded as but a ﬁrst step in the study of the effects of area changes on the ﬂow. Effects of viscosity in this ﬂow are the subject of other chapters, Chaps. 1, 2. 9.2.2 Scope of the Present Work Euler’s equations of motion for the steady axisymmetric ﬂow of a ﬂuid with constant density are therefore at the foundation of our study. Our aim is to extract, based on

9.3 Mathematical Formulation of the Problem

153

this foundation, salient characteristics of the ﬂow at the outlet section for a given distribution of the axial and azimuthal velocity components at the inlet when there is a change in area/radius of the ﬂow passage. Literature covering studies of this problem is indeed very extensive, particularly when the ﬂow is irrotational. The ﬂow under study here is, although inviscid, not irrotational, and many methods have been developed to solve the Euler’s equations for this class of problems too. These methods are invariably numerical, and results of numerical methods inherently have the character of being solutions to approximations in some sense. Therefore, in order to judge the validity of the results, it is highly desirable that they be subjected to one or more tests. One such test is outlined in the following and will be discussed in more detail later: It can be derived from the differential equations of motion that two ﬂow quantities of signiﬁcance, viz. the total head and the moment of the azimuthal velocity, remain constant on streamsurfaces. Since this physical constraint follows from the differential equations themselves, compatibility of the numerical results with this condition of invariance is a test for the validity of the approximations inherently present. Since, to the author’s knowledge, none of the methods incorporate such a check, it appeared meaningful to develop one in which such a test is an integral part of the method, and this is the course charted out for the present work.

9.3 Mathematical Formulation of the Problem A sketch of the ﬂow problem addressed is shown in Fig. 9.1. It is the internal ﬂow in a passage of circular cross section with radius R(x). R1 and R2 denote the radii at the inlet and outlet sections respectively. The analysis to follow does not place any restriction on the manner of change of the cross-sectional area as long as its character of rotational symmetry is maintained and the rate of change of the radius is only gradual. Contraction and expansion of the passage area are both permissible \ <

\= constant R1 = a r

\ \

R(x)

R2 = b

r* x

Inlet section

Outlet section

Fig. 9.1. Geometric conﬁguration.

154

9 Effect of Area Changes in Swirling Flow

in principle for our analysis, but our interest lies primarily in the case of contraction, R1 > R2 , since this is the case most frequently met with in ﬂow metering practice. The proﬁles of the axial and azimuthal velocity components at the inlet cross section may be speciﬁed arbitrarily, again with the restriction that they be axisymmetric. The ﬂow quantities sought are the axial and the azimuthal velocity proﬁles at the exit cross section. These shall be obtained as a solution of the Euler’s equations of motion satisfying the boundary condition that the speciﬁed geometry R(x) be a stream surface. With the x-axis chosen to coincide with the axis which is the main direction of the ﬂow, the equation of continuity and the Euler’s equations of motion governing the steady axisymmetric ﬂow of a ﬂuid at constant density are as follows, see eg. Batchelor (1967), Schlichting and Gersten (2000): ∂(rux ) ∂(rur ) + =0, ∂x ∂r 1 ∂p ∂ux ∂ux + ur =− , ∂x ∂r ρ ∂x

(9.2)

u2φ 1 ∂p ∂ur ∂ur + ur − =− . ∂x ∂r r ρ ∂r

(9.3)

ux

ux

(9.1)

The equation of continuity, (9.1), may be satisﬁed by introducing a stream function ψ deﬁned through (9.4) below: 1 ∂ψ 1 ∂ψ ; ur = − . (9.4) r ∂r r ∂x An important result that is derivable from the above (9.1), (9.2), (9.3), (9.4), is that the two quantities, the total head H and the moment of azimuthal velocity u φ , deﬁned as, p 1 C dC 1 2 ux + u2r + u2φ + = u2x + dr , (9.5) H= 2 ρ 2 r2 dr ux =

and C = ruφ ,

(9.6)

remain constant along streamsurfaces ψ = constant, i.e., H = H(ψ) and C = C(ψ). For a derivation of this result the reader is referred to standard literature on ﬂuid dynamics, eg. Batchelor (1967). Pressure may be eliminated from the equations, (9.2) and (9.3) to yield an equation for the azimuthal component of the vorticity, in which the deﬁnition of ψ may be used to obtain a single governing partial differential equation for ψ. This is as follows (see Eq. 7.5.11 in Batchelor (1967)): dC(ψ) dH(ψ) ∂ 2 ψ 1 ∂ψ ∂ 2ψ = r2 − C(ψ) . + − 2 ∂x ∂r2 r ∂r dψ dψ

(9.7)

9.3 Mathematical Formulation of the Problem

155

The velocity ﬁeld satisfying the equations (9.1), (9.2), (9.3) is required to fulﬁll only the kinematic boundary condition of the velocity component normal to the bounding surface being zero. For (9.7) this may be written as follows: r = R(x) : ψ = Ψ .

(9.8)

When the rate of change of the cross section radius is small, the derivative with respect to x can be neglected compared to the other terms and we get the following ordinary differential equation for ψ(r), (9.9), as an approximation. This approximation then captures the salient features of the ﬂow at an arbitrary station x where the cross-sectional radius is R(x). The distribution of H and C at the initial station of cross-sectional radius R1 is to be regarded as given, and hence known. The approximation then obeys the following equation: dC(ψ) dH(ψ) d2 ψ 1 dψ = r2 − C(ψ) , − 2 dr r dr dψ dψ

(9.9)

which is the same as for the problem of cylindrical ﬂows dealing with transition from one long cylindrical section to another of different radius. It has been treated in Batchelor (1967), Chap.7.5. If the above equation is understood as an approximation for the ﬂow at any value of x in a passage of circular cross section with a gradual change in radius, the x-dependence of the ﬂow is visible only in a parametric form through the outer boundary condition which remains the same as in (9.8). The difﬁculties of obtaining a solution of (9.9) for arbitrary distributions of H and C at the inlet cross section are well-known, see, e.g., Batchelor (1967). However, the particular case of a constant axial velocity U and solid body rotation (constant angular velocity Ω) at the inlet section turns out to be relatively simple and the solution for this problem is given in Batchelor (1967). In view of the importance of this special case for the more general problem to follow, we summarise in the subsection below the outlines of this special case. 9.3.1 Solution for the Special Case of Homogeneous Axial Velocity and Solid-Body Rotation at Inlet When the ﬂow conditions at inlet are such that the axial velocity proﬁle is homogeneous with a value U and the swirl is a solid-body rotation with angular velocity Ω, H and C may be written as linear functions of ψ, viz. C= and

2Ω ψ, U

(9.10)

1 2 2Ω 2 U + ψ, (9.11) 2 U and the solution to (9.9) is expressible through Bessel’s functions. We need this solution for later reference and summarise its essentials here for the reader’s convenience. In our present notation the solution may be written as follows: H=

156

9 Effect of Area Changes in Swirling Flow

ψ=

1 2 U r + rF (r) , 2

where F (r) is a solution of the Bessel’s equation of order unity given by 1 dF 1 d2 F 2 + k − 2 F =0, + dr2 r dr r

(9.12)

(9.13)

where

2Ω . (9.14) U The solution for F satisfying the boundary conditions of the problem may be written as F = AJ1 (kr) + BY1 (kr) , (9.15) k=

where J1 and Y1 denote Bessel functions of the ﬁrst and second kind respectively, and A and B are constants related to the radii at the inlet and outlet sections. This solution corresponds to the axial and the azimuthal velocity distribution at the outlet section given by the following expressions: ux =

1 ∂ψ = U + AkJ0 (kr) + BkY0 (kr) r ∂r

(9.16)

and

2Ωψ C = = Ωr + kAJ1 (kr) + kBY1 (kr) . (9.17) r Ur The solution in (9.16), (9.17) is applicable to the more general geometrical conﬁguration of the swirling ﬂow in the annulus between concentric passages of circular cross section. In the absence of the inner passage the constant B equals zero. We complete the summary of this simple case by giving the expressions for the axial and azimuthal velocity components ux and uφ which will be needed for extension to the more general case to be dealt with shortly. Evaluating the constants A and B in the expressions (9.16), (9.17) for the ﬂow geometry of transition from a radius a upstream to b downstream, and a homogeneous axial velocity at the value U and a homogeneous angular velocity (solid body rotation) of value Ω, we get: 2 a kbJ0 (kr) ux =1+ , (9.18) − 1 U b2 2J1 (kb) uφ =

and

uφ =1+ Ωr

a2 −1 b2

bJ1 (kr) . rJ1 (kb)

(9.19)

An important property of this result, which will be of use later, is that u φ at outlet scales with Ω, the angular velocity of the swirling motion at inlet.

9.3 Mathematical Formulation of the Problem

157

9.3.2 The General Case of Arbitrary Proﬁles of Axial and Azimuthal Velocity Components at Inlet The main source of difﬁculty in solving the equations of motion for the boundary conditions speciﬁed of the present problem is traceable to the vorticity at the inlet section. This difﬁculty is more readily visible in the vorticity equation that does not contain the pressure, viz. (9.7) than in the original form with the “primitive variables” (9.1), (9.2), (9.3). The left-hand side of the vorticity equation (9.7) is the Laplace operator for the stream function which would have been equal to zero if the ﬂow had been irrotational. We may recall that, in this case, the coupling between the kinematic and the dynamic problems is only in one way, i.e., the kinematic and the dynamic problems may be solved successively in this order, since the kinematic problem of determining the velocity ﬁeld is not coupled with the dynamic problem. Pressure is then determinable from the Bernoulli equation after the velocity ﬁeld is obtained. The rotational nature of the ﬂow destroys this simplicity and the kinematic and dynamic problems then require to be solved simultaneously. The right-hand side of (9.7) illuminates the nature of the coupling involved. The mutual coupling between the kinematic and the dynamic problems is present even in the absence of swirl (Ω = 0) due to the total head H being a function of ψ, but the presence of swirl complicates the problem further. The difﬁculty is not removed if attention is restricted to gradual changes of the cross section alone, i.e., if the approximation (9.9) is regarded as the governing equation instead of (9.7). For arbitrary distributions of the axial and azimuthal velocity components, u x ,uφ , at the inlet cross section, it will not be possible to express H and C in terms of ψ in an analytical form. As a matter of fact it may not be feasible to ﬁnd convenient analytical expressions for describing the proﬁles of u x and uφ at the inlet section as functions of r, and even then only with difﬁculty. Keeping the desirability of application of this study for questions arising in metering in mind, it is more reasonable to work on the hypothesis that ux and uφ will be available only in tabulated form. The table can be thought of as originating from experimental data. The data in tabular form may also be extracted from another kind of source, say, from post-processed numerical solutions of the equations of motion for the swirl-generating element of the installation that preceeds the pipe, eg. the elbow or the compressor. It may therefore be expected that one has to resort to numerical methods for obtaining the solution to the problem on hand. In the present state of algorithm development, methods to solve the equations (9.1), (9.2), (9.3) or (9.7) are indeed available, see eg. Chap. 11. Many of these methods are available in the form of published algorithms paving the way for the reader to choose one that suits his or her purpose best. However, by their very nature, numerical methods involve some kind of approximation, a matter of fact that, in the author’s opinion, should not be lost sight of when judging the validity of the results. It is therefore desirable to keep a check on the validity of the results whenever this is feasible. Since the numerical results are solutions of approximations to the differential equations, a suitable check for their validity would be if they satisfy the invariance constraints that is derivable from the differential equations directly, see

158

9 Effect of Area Changes in Swirling Flow

eg. Ferziger and Peric (2002). In devising such a test, it is meaningful to critically examine the underlying principles on which the numerical method intended for use rests. We observe in this context that there is a common principle at the roots of most of the numerical methods in current use for which algorithms are readily available. It is that they work with x and r as the independent variables, or r alone (in (9.9)), as the case may be. Choices for the dependent variable are several in number, and so are the approximations for the dependent variable underlying the discretisation process. But regardless of these steps, it is inherent in this principle of approach that, if the ﬂow quantities (ux , uφ , p) or ψ are obtained as functions of r at some x, a check on their compatibility with the solutions that would be obtainable from the differential equations would involve examining if H and C remain invariant along streamsurfaces ψ = const.. A test for this invariance is, to the author’s knowledge, not incorporated as a routine in any of the algorithms hitherto published, neither can such a test be added as a simple and straightforward extension to the algorithms already available. For application to the class of problems in question, preference for an algorithm that incorporates the invariance test is evident, and the revised approach presented herein is to be understood as one such. The principle underlying the present approach is that instead of trying to solve for ψ as a function of r and checking if it satisﬁes the physical constraints imposed, we interchange the dependent and independent variables and look for r as a function of ψ. In this approach, if the radius r is obtained for a certain streamsurface ψ, the azimuthal velocity uφ follows from C, which, being constant on the streamsurface, has the same value as at the initial station. Further, since H also stays constant on the streamsurface at its initial value, ux follows from Bernoulli’s equation in the form (9.5). In a method based on this approach the invariace property of H and C remaining constant on streamsurfaces is therefore integrated into the algorithm. The starting point in this approach is therefore (9.9), with the dependent and independent variables interchanged. This step leads to: d2 r r 2+ dψ

dr dψ

2 +r

3

dr dψ

3

dH(ψ) −r dψ

dr dψ

3 C(ψ)

dC(ψ) =0 dψ

(9.20)

The wall boundary condition for (9.20) is the same as before, viz. (9.8), of course with the understanding that r is the dependent variable and ψ the independent variable. A numerical method suited to solve this boundary-value problem is the shooting method. However some care is necessary in handling the boundary condition on the axis when r is regarded as the dependent variable for the reason to be outlined below, and the method of solution is required to pay due regard to this aspect of the problem. The Boundary Conditions on the Axis and at the Wall For the case of present interest, where the axial velocity u x and the azimuthal velocity uφ are permitted to have arbitrary distributions along the radius at the initial station, it is not overly restrictive to hypothesize that they are regular and so can be approximated in the following way in the region around the axis r = 0 :

9.3 Mathematical Formulation of the Problem

159

u x = U0 + P r 2 + o r 2 ,

(9.21)

uφ = Ω0 r + o(r) .

(9.22)

and The expressions (9.21) and (9.22) convey that in a small neighbourhood of the axis the axial velocity is describable by a parabola and the azimuthal velocity corresponds to solid-body rotation with the angular velocity Ω0 . Equation (9.21) implies that the streamfunction ψ according to (9.4) and its derivative may be written around the axis as follows: P r4 U0 r 2 + + o r4 , (9.23) ψ= 2 4 and dψ = U0 r + P r 3 + o r 3 . (9.24) dr On interchanging the dependent and independent variables, the expression for dψ/dr according to (9.24) corresponds to the following expression for dr/dψ in the region around the axis: 1 Pr dr = − 2 + o(r) , dψ U0 r U0

(9.25)

which clearly shows the singular nature of this derivative on the axis. This is only a different manifestation of the singularity on the axis that is known to arise when polar coordinates are employed for computation. It necessitates exclusion of the origin r = 0 itself from numerical computations. Since the governing equations for our work are the Euler’s, and not the Navier– Stokes, equations, only the kinematic boundary condition can be imposed at the wall. In terms of the stream function, this means that only the value of ψ at the wall can be prescribed. It is in order to note in this context that, when the shooting method is employed, and integration of (9.20) is to be carried out from the wall towards the axis, the starting value for ux at the wall (r > 0) has to be prescribed to be different from zero, since otherwise the derivative dr/dψ = 1/(rux ) becomes inﬁnite. Since (9.20) is to be satisﬁed at the inlet section too, the axial and azimuthal velocity proﬁles at the inlet section, and through them H and C, may be prescribed arbitrarily but u x at the wall should not be zero. This is tantamount to admitting slip at the wall which is permissible when working with Euler’s equations. Method of Solution A suitable method of solution to the problem on hand is the shooting method. It is straightforward in principle, and reduces solving the boundary-value problem to two simple more elementary steps, viz. solving an initial-value problem for a range of guessed initial values, followed by ﬁnding the correct initial value through ﬁnding the zero of a suitably deﬁned function quantifying the error. Accordingly, the ﬁrst

160

9 Effect of Area Changes in Swirling Flow

step is carrying out integration of (9.20) from the wall towards the axis, say with a Runge–Kutta method, with the value ψ = Ψ given at the boundary, for several guessed values of dr/dψ = 1/(rux ) at the wall. It is in order to recall here that ux at the wall cannot be prescribed to be zero. This step yields values of the unknown variable r = r∗ at a certain value of the independent variable ψ = ψ ∗ , which is chosen to lie close to the axis but be different from zero. We thus have r ∗ as a function of the derivative dr/dψ at the wall. The second step is ﬁnding the zero of an error deﬁned through the difference between the value of this (discrete) function and the “correct” value of r belonging to ψ ∗ . The “correct” value of r that belongs to ψ ∗ may be obtained in the following way: Since the streamsurface ψ ∗ lies close to the axis, the axial velocity proﬁle at the inlet section may be regarded as uniform with the value U0 in (9.21), and the azimuthal velocity proﬁle at this section a linear function of the radius corresponding to solidbody rotation with an angular velocity Ω0 in (9.22). With this approximation, the axial and azimuthal velocity components at any downstream section x are distributed according to (9.18) and (9.19) respectively and the relation between ψ ∗ and r∗ may be written as follows, see. (9.26): ∗

r

ψ =

∗

U0 r ∗ 2 + rux dr = 2

0

r

∗

r 0

r0∗2 −1 r∗2

Ω0 r∗ J0 (2Ω0 r) dr J1 (2Ω0 r∗ )

(9.26)

The above expression is not very convenient for numerical evaluation of r ∗ at a given value of ψ ∗ . But, if the arbitrarily speciﬁed distribution of the axial velocity component at the inlet section departs only mildly from the constant value U 0 , i.e., if the inhomogeneity is small, it is straightforward to obtain r ∗ for a chosen ψ ∗ in terms of the local radius of the circular passage R(x) through integration of (9.18). This is as follows: ∗2 R12 kR r ∗2 + − 1 r (9.27) ψ ∗ = U0 2 R2 (x) 4J1 (kR)

9.4 Results and Discussion The equation (9.20) has been solved by the method described for a number of cases, and we present in the following only an illustrative sample. However, at ﬁrst we wish to subject our method to a test and examine if it recovers the known results for the standard case of homogeneous distributions of both the axial and the angular velocity at the inlet section. 9.4.1 Veriﬁcation of the Method The solution of (9.20) for u and w/r when conditions of a homogeneous axial velocity and solid body rotation are prescribed at the inlet are compared with Batchelor’s

9.4 Results and Discussion

161

1.2 1

r

0.8 0.6 0.4 0.2 0 0

0.5

1

1.5

2

1.5

2

u

1.2 1

r

0.8 0.6 0.4 0.2 0 0

0.5

1 w/r

Fig. 9.2. Effect of contraction for homogeneous inlet conditions: Contraction ratio R2 /R1 = b/a = 0.9. Proﬁles at inlet: (+ + +) u ≡ 1, w/r ≡ 1. Proﬁles at outlet: (—–) analytical solution according to (9.18), (9.19); (2 2 2) numerical solution of (9.20).

solution for this problem, viz. (9.18), (9.19), for a contraction ratio R 2 /R1 = 0.9 in Fig. 9.2. The comparison shows that our method captures the results of the standard case satisfactorily. This relatively simple example already points at an important qualitative character of the effect of an area change on the ﬂow, viz. that u and w/r do not any longer remain constant along a radius downstream of a constriction. 9.4.2 The Effects of Passage Contraction Figure 9.3 shows proﬁles of u and w/r caused by a passage contraction when u departs from a constant but w/r remains a constant at the inlet section. The nonhomogeneous axial velocity proﬁle at inlet in this example is prescribed to be “mildly parabolic”, i.e., a parabola of mild curvature superimposed on a homogeneous axial

162

9 Effect of Area Changes in Swirling Flow 1.2 1

r

0.8 0.6 0.4 0.2 0 0

0.5

1

1.5

2

1.5

2

u

1.2 1

r

0.8 0.6 0.4 0.2 0 0

0.5

1 w/r

Fig. 9.3. Effect of contraction for nonhomogeneous inlet conditions: Contraction ratio R2 /R1 = 0.9. Proﬁles at inlet: (+ + +) u = 0.95 + 0.1(1 − r 2 ), w/r = 1. Proﬁles at outlet: ( ) numerical solution of (9.20).

velocity proﬁle. The change in character of the distributons of u and w/r at outlet are evident here too. A cursory inspection of the plots of u and w/r in Figs. 9.2 and 9.3 already draws attention to an outstanding qualitative feature of the effect of a constriction on the distribution of the ﬂow quantities. It is that the shape of distribution of the ﬂow quantities is no longer preserved. The character of the change is more pronounced in the distribution of the angular velocity w/r than of the axial velocity proﬁle u. Not only is the homogeneous distribution of w/r at inlet, which is solid-body rotation, destroyed as such by the area constriction, but also the nature of the change when the axial velocity proﬁle at inlet is nonhomogeneous departs from that when u is homogeneous. The signs of the curvature of the proﬁles of w/r at the outlet section are different in the two cases.

References

163

The above examples are indicative of the sensitive dependence of the axial and the azimuthal velocity proﬁles at the exit section to the proﬁles at the inlet section. Consequently, knowledge of the proﬁle details at the exit section presupposes knowledge of these details at the inlet section which, in most ﬂow metering applications, is rarely, if ever, available. Therefore, conclusive statements of a general nature on the velocity proﬁles perceived by the metering device do not seem to be within reach at present. However, when the inhomogeneity in the proﬁles of both the axial velocity and the angular velocity at the inlet section is small, analysis based on the dynamics of ﬂuid ﬂow can be taken a step further along lines to be described below. The more convenient starting point for such an analysis is (9.9). We introduce for this purpose a small parameter characterizing the inhomogeneity, in which, for instance may be the maximum departure of the actual proﬁles from the homogeneous case. We may then write the axial and azimuthal velocity proﬁles at the inlet section as follows: ux = UH (1 + in Fu (r)) ,

(9.28)

uφ = ΩH r(1 + in Fφ (r)) .

(9.29)

and For the case of solid-body rotation, Fφ (r) is a constant, and in may be set equal to 0. It is then straightforward to verify that ψ, H, C, dH/dψ and dC/dψ all assume the same form to the order O(in ) at inlet. Solutions for (9.9) may then sought in the form of asymptotic expansions in terms of the parameter in , from which it follows that the velocity proﬁles at exit are also representable through asymptotic expansions of the same form. The conclusion that may be drawn herefrom is that, to a leading approximation in in , the effect of an area change on the velocity distribution scales in the same mannor as in the ﬂow with homogeneious inlet conditions.

References Batchelor GK (1967) An Introduction to Fluid Dynamics. Cambridge University Press, Cambridge UK Brown GL, Lopez JM (1990) Axisymmetric vortex breakdown, Part 2: Physical mechanisms. J Fluid Mech 221:553–576 Buntine JD, Saffman PG (1995) Inviscid swirling ﬂow and vortex breakdown. Proc. Royal Soc. London A 449:139–153 Escudier M (1988) Vortex breakdown: Observations and explanations. Prog. Aerospace Sc. 25:189–229 Faler JH, Leibovich S (1978) An experimental map of the internal structure of a vortex breakdown. J Fluid Mech. 86:313–335 Ferziger JH, Peric M (2002) Computational Methods for Fluid Dynamics. Springer, Heidelberg Gallaire F, Chomaz J-M (2003) Instability mnechanisms in swirling ﬂows. Physics of Fluids 15:2622–2639 Marshall JS (2001) Inviscid Incompressible Flow. John Wiley, New York

164

9 Effect of Area Changes in Swirling Flow

Rusak Z, Wang S, Whiting CH (1998) The evolution of a perturbed vortex in a pipe to axisymmetric vortex breakdown. J Fluid Mech. 366:211–237 Schlichting H, Gersten K (2000) Boundary Layer Theory. Springer, Heidelberg Wang S, Rusak Z (1997) The dynamics of a swirling ﬂow in a pipe and transition to axisymmetric vortex breakdown. J Fluid Mech 340:177–223

10 Errors of Turbine Meters Due to Swirl Venkatesa Vasanta Ram Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

This chapter addresses the question of the error in metering of a ﬂow with a turbine ﬂow meter caused by swirl in the oncoming ﬂow. The problem is approached by analyzing the signal-generation mechanism of a turbine ﬂow meter from a fundamental ﬂuid mechanics point of view when the oncoming ﬂow shows departures from its calibration state. The fundamental physical mechanism in question is the driving moment experienced by the rotor of the turbine ﬂow meter due to the aerodynamic forces on the rotor blading. Swirl in the oncoming ﬂow belongs in this scheme to one class of possible disturbances to the calibration ﬂow. The study brings out the salient parameters that govern the error in metering due to swirl, and the pattern of error dependance on swirl strength. Notation x, r, φ ρ, ν ux , uφ vx , vφ R, D Ri , Ro q(r) Q ˆ Q(r) Uref ΩR Re

co-ordinate system: axial, radial and azimuthal directions respectively density and kinematic viscosity of the ﬂowing ﬂuid x and φ velocity components at inlet section to turbine ﬂow meter, section A-A in Fig. 10.1 x and φ velocity components at inlet section to rotor blading, section B-B in Fig. 10.1 pipe radius and diameter respectively at inlet section to ﬂow meter, Fig. 10.1, D = 2R hub and tip radii of rotor rΩRc deﬁned in (10.8), q = vxc volume ﬂow rate through the ﬂow meter deﬁned through (10.14) Q bulk velocity, reference velocity, deﬁned through U ref = πR2 angular velocity of rotor Uref D Reynolds number, Re = ν

166

10 Errors of Turbine Meters Due to Swirl

Ro S Q αbl (r) Lbl (r) Tbr Taero Fmeter1 , Fmeter2 FT Bbr

ωS R Uref parameter characterizing swirl strength, deﬁned in(10.21) error in metering of the ﬂow, deﬁned in (10.39) blade angle with respect to rotor axis at radius r, Fig. 10.2 chord of blade at radius r, Fig. 10.2 braking torque on rotor of turbine ﬂow meter driving (aerodynamic) torque on rotor of turbine ﬂow meter functions deﬁned through (10.15), (10.16) function deﬁned through (10.20) deﬁned in (10.34) Rotation number, Ro =

Subscripts c 1 bl R S ref meas

denotes calibration state, e.g., vxc , ΩRc , αc , Tbrc , Taeroc denotes departure from calibration state, e.g., ΩR1 , Q1 , α1 , Tbr1 , Taero1 refers to rotor blade, e.g., αbl , Lbl refers to ﬂow meter rotor, e.g., ΩR refers to swirl or swirl generator, e.g., S , ωS denotes reference, e.g., Qref , Uref denotes measurement in Qmeas

10.1 Introduction The turbine ﬂow meter is a type of ﬂow metering device very commonly adopted in engineering practice. This type of ﬂow meter is often preferred to alternatives in many engineering applications, the reason for which is that it offers the advantages of relative simplicity in installation and operation combined with ruggedness and acceptable accuracy for the purpose of the application in question, see e.g., Baker (2000). The turbine ﬂow meter shares a feature common to metering devices in general in that it needs calibration. Calibration is generally done in the fully developed pipe ﬂow that is free of swirl. Swirl in the ﬂow of measurement causes relevant details of the ﬂow to depart from the calibration state, and this in turn leads to errors in metering. The task of estimating this error calls for an analysis of the physical mechanisms involved in the conversion of the ﬂow rate into the metering signal under measuring conditions. A closer examination of this mechanism for a turbine ﬂow meter reveals what may be regarded as a salient characteristic of this metering device not shared in common by most of the other types. It is that the error is sensitive to the direction of swirl in the oncoming ﬂow. The purpose of the present section is to present an analysis of the characteristics of this error together with a comparison of the same with experiments.

10.1 Introduction

167

A cursory examination of the operating principle of the turbine ﬂow meter is sufﬁcient to hint at the strong similarities that exist between the questions arising in the context of the turbine ﬂow meter and those in propellers and/or ducted fans. Propellers and ducted fans belong to a classical branch of study in ﬂuid mechanics in which extensive body of literature has been accumulated, see e.g., Leishman (2000), and it is only too natural to look for ways of carrying over results already established and proven, to the ﬂow meter problem. However, a closer examination of the two problems shows that, despite the similarities that undoubtedly exist, there is a significant difference between the two that stands in the way of results from the classical studies being taken over in a straightforward manner to the ﬂow meter. The main difference is that the ﬂow meter is exposed, both in calibration and measurement, to rotational ﬂow, whereas most of the classical studies of propellers and ducted fans are centered around irrotational ﬂow. Thus, although the basic ideas and approach in the classical studies can be carried over to the ﬂow meter problem, differences in results are to be expected, warranting a separate study of the ﬂuid mechanics of the ﬂow meter. Carrying out such a study is the object of the present work. The standard work of Baker (1993) is a survey of techniques for ﬂow measurement in general, and the reader is referred to this work for a thorough review of the theoretical foundations and operating principles of the turbine ﬂow meter too. The physical principle on which the turbine ﬂow meter rests for its operation is the balance of torque on its rotor. A quantitative formulation of this physical principle leads to the result that the angular velocity of the rotor is decided by the balance between the driving torque due to the aerodynamic forces on the rotor blading and the braking torque (or drag torque) exerted on the rotor. Published literature on the turbine ﬂow meter identiﬁes four sources of drag, viz. the bearing drag, the hub disk friction drag, the tip clearance drag and the hub ﬂuid drag, see Baker (2000). It is a reasonable hypothesis to suppose that swirl in the oncoming ﬂow leaves the physical mechanisms behind the drag torque from these four sources unaffected. So it appears justiﬁable that an analysis of the error of the turbine ﬂow meter caused by swirl be founded, to start with, on an examination of the effect of swirl in the oncoming ﬂow on the aerodynamic torque exerted on the rotor blading, and this is the subject in focus in the present study. Scope of the Present Work The starting point for this examination is the expression for the aerodynamic torque on the rotor under measurement conditions when swirl may be present in the oncoming ﬂow. The relation between the actual ﬂow rate and the angular velocity of the ﬂow meter rotor under measurement conditions follows from this expression. This expression deﬁnes the calibration curve of the ﬂow meter which is the relation between the ﬂow rate and the output signal when calibration conditions exist, i.e., when the oncoming ﬂow is free of swirl and the axial velocity proﬁle corresponds to the calibration state. The next step is the derivation of an expression describing the departure of this relation between the ﬂow rate and the output signal (rotor angular velocity) when the velocity proﬁle of the oncoming ﬂow is disturbed from its calibra-

168

10 Errors of Turbine Meters Due to Swirl

tion state. Swirl in the oncoming ﬂow is only one among the several possible classes of disturbance, see Chaps. 2 and 3. Suitable parametrization of this speciﬁc class of disturbance leads to an expression relating the ﬂow meter reading error and the swirl parameter. We then describe an experiment conducted in a specially designed facility in which a turbine ﬂow meter is exposed to an oncoming ﬂow into which swirl of known strength and distribution is purposely introduced. The experiment yields the error in the ﬂow rate “read” by the turbine ﬂow meter due to swirl in the oncoming ﬂow. The experimental outcome is set against the analysis over a range of the swirl parameter. The outline of this chapter is as follows: In Sect. 10.2 we derive the expression for the aerodynamic torque on the rotor under conditions when swirl may be present in the ﬂow. In Sect. 10.3 we describe an experiment and present the data therefrom reduced in a form suitable for setting against the theory of Sect. 10.2. The discussion that follows in Sect. 10.4 then leads to the conclusions drawn in Sect. 10.5.

10.2 Physical and Mathematical Background Figure 10.1 is a sketch showing the main components of a turbine ﬂow meter relevant for the present study. The physical law underlying the relation betwen the ﬂow rate Q and the output signal ΩR of the ﬂow meter is the balance of torque on the rotor of the ﬂow meter when this is exposed to stationary ﬂow conditions. This may be written as follows: (10.1) Tbr = Taero , where Tbr and Taero are the braking and driving torques respectively. Since the ﬂow meter is neither driven by any external power source, nor is it delivering power, the driving torque is only due to aerodynamic forces on the rotor blading, hence the notation Taero for the driving torque.

Fig. 10.1. Turbine ﬂow meter components

10.2 Physical and Mathematical Background

169

10.2.1 Aerodynamic Torque on the Turbine Flow Meter Rotor The velocity and force triangles on a blade element at a radial location r are shown in Fig. 10.2. Figure 10.2a depicts these triangles under calibration conditions whereas Fig. 10.2b refers to conditions when the oncoming ﬂow contains swirl. The expression for the aerodynamic torque on the rotor may be obtained by integrating the contribution to the torque from the aerodynamic force on a blade element from the hub to the tip. From dimensional considerations the magnitude of the aerodynamic force vector dFaero on the blade element between the radii r and r + dr may be written as follows: |dFaero | = ρv 2 Lbl drC ,

(10.2)

where ρ is the density of the ﬂuid ﬂowing through the meter, L bl the local chord of the rotor blading, v the velocity “perceived” by the blade at the radial location r, see Fig. 10.2, and C a dimensionless coefﬁcient that depends upon the blade form and other geometrical factors entering the design of the meter. The azimuthal component of this force, denoted dFaeroφ is then given by dFaeroφ = ρv 2 Lbl drC cos (αbl − α) ,

(10.3)

where αbl is the angle the blade makes with the rotor axis and α the angle of incidence the velocity vector v makes with the blade, see Fig. 10.2a, b. The velocity vector v is as “perceived” by the blading, i.e., it is in a frame of reference that is rotating about the ﬂow meter axis with the angular velocity of the rotor, Ω R , and α is measured from the direction of zero lift of the blade. It is reasonable to regard this vector v as one with negligible radial component even when swirl is present in the oncoming ﬂow. The axial and azimuthal components of v are therefore as shown in Fig. 10.2a, b. The torque exerted by the aerodynamic force on the rotor blading may then be obtained by integrating rdFaeroφ across the blade from Ri to Ro . The expression for the aerodynamic torque Taero is then: Taero

Ro = ρ rv 2 Lbl C cos (αbl − α) dr

(10.4)

Ri

In the further course of the analysis to follow we work with the hypothesis that the force vector dFaero is perpendicular to the velocity vector v , which is tanamount to postulating the aerodynamic torque to be only due to the lifting force on the blade. For a discussion of the meaning and implications of this hypothesis the reader is referred to standard texts on fundamentals of ﬂuid mechanics, e.g., Batchelor (1967), Marshall (2001), Schlichting and Truckenbrodt (1962). The hypothesis is justiﬁable when the ﬂow meter operates at high Reynolds numbers, which is mostly the case, and boundary layers remain thin and attached to the blading. We also further set the lift to be proportional to the angle “perceived” locally by the blade section, i.e., C ∼ α. So we may write C = N CL α where N is the number of blades on the

170

10 Errors of Turbine Meters Due to Swirl

Fig. 10.2. Velocity and force triangles

10.2 Physical and Mathematical Background

171

rotor and CL in (10.5) is the lift coefﬁcient of a single blade, of course in a cascade arrangement. Taero in (10.4) is then given by Taero

Ro = N ρ CL rv 2 Lbl α cos αbl dr.

(10.5)

Ri

The step leading to the expression in (10.5) from (10.4) involves the approximation for small values of α obtainable on use of the standard trigonometric relationship between cos (αbl − α) and circular functions of αbl and α. Furthermore, the number of blades on the rotor has been explicitly accounted for through N which is hidden in the dimensionless coefﬁcient C in (10.4). For obvious reasons CL may depend upon the radial location r, hence its position under the integral sign. It is useful at this stage to introduce the relation between the angle α and ﬂow parameters entering the problem which are the following: • the axial velocity component vx (r), • the azimuthal velocity component in the laboratory coordinate system which may be written in terms of the local angular velocity of the ﬂuid, ω(r), as rω(r), and • the angular velocity of the ﬂow meter rotor ΩR . This relation is: tan(αbl − α) =

r (ΩR − ω) , vx

(10.6)

which may also be written using standard trigonometric identities as follows: vx , cos αbl cos α + sin αbl sin α = 2 vx2 + r2 (ΩR − ω)

(10.7)

It is appropriate at this juncture to pause brieﬂy to reﬂect upon a certain point in the hypothesis that touches a fundamental aspect of ﬂuid dynamics and is implicit when writing (10.5, 10.6, 10.7). It is that the vector v “perceived” by the blade element cannot be set to be just the ﬂow “far upstream”, merely seen from an appropriate coordinate system. More properly, it should be the ﬂow far upstream modiﬁed by the vortex system in the wake of the rotor, whereby the question of resolving the effective angle of incidence α acquires key importance. The physical reasoning behind this conjecture is essentially the same as in classical wing theory, where the effective angle of incidence that accounts for the lift felt at a certain spanwise section of a wing of large but ﬁnite span is given by the ﬂow direction far ahead of the wing corrected by the downwash induced by the vortex system in the wake of the lifting surface. The observation known from classical wing theory, viz. that the downwash depends upon details of the lift distribution along the span of the lifting surface, holds for the problem of current interest too. However, in the present context, which is the task of assessing the effects of swirl on ﬂow measurements with turbine ﬂow meters, carrying over this concept to the problem on hand through accounting for the effect of this vortex system in general terms for turbine ﬂow meters does not seem to be

172

10 Errors of Turbine Meters Due to Swirl

warranted. The reason is that the answer would depend upon such design details of the ﬂow meter that are rarely, if ever, available to the user, hence tending to make the exercise futile. We therefore chart out a course for our present study in which the induced velocity ﬁeld of the vortex system in the wake of the rotor is divided into two further contributions. These are: 1) the induced velocity ﬁeld when the ﬂow meter is exposed to the calibration ﬂow, and 2) that when swirl or any other kind of disturbance is present in the ﬂow of measurement. The meaning of this formulation, as it will shortly be seen, is that while enabling the induced velocity ﬁeld in the calibration ﬂow to be accounted for, that due to swirl in the oncoming ﬂow or another kind of disturbance, is ignored. This approach is analogous to an established method of handling the effects of rotation in the free stream in inviscid airfoil theory, in which it is stipulated that there be no additional circulation due to rotation in the free stream e.g., Van Dyke (1975). In the present context the justiﬁcation for this approach is that the result is applicable more generally to turbine ﬂow meters. A more rigorous treatment of the complete problem would call for an approach through solving the full equations of motion with the appropriate boundary conditions. For an example of this approach for a particular ﬂow meter the reader is referred to Chap. 11. Calibration Conditions Under calibration conditions the axial velocity proﬁle at the inlet to the ﬂow meter, this is section A-A in Fig. 10.1, is the fully developed ﬂow in a pipe which we denote as uxc , and the ﬂow is free from swirl. In turbulent ﬂow, the shape of the proﬁle u xc depends upon the Reynolds number, see Chap. 1 for the nature and details of this dependence. It follows therefore that, for a certain given Reynolds number, there will be, under calibration conditions, a certain velocity distribution u xc at the inlet section of the ﬂow meter (section A-A of Fig. 10.1) to which there will correspond a certain distribution vxc at the inlet section to the rotor blading (section B-B of Fig. 10.1) too. The latter, vxc , depends upon the geometrical details of the meter and hence may differ from design to design. It is reasonable to presume that aspects of meter design, say guide vanes, do not introduce a non-zero azimuthal velocity at the section B-B, so we may work with vφc = 0. The velocity triangle for this case is as shown in Fig. 10.2a. The relation sought involves the ratio (rΩRc )/(vxc ) for which it is convenient to introduce an abbreviation q in terms of which (10.6, 10.7) then reduce to the following, where the subscript c denotes the calibration state: tan(αbl − αc ) = and

rΩRc = q(r), vxc

1 . cos αbl cos αc + sin αbl sin αc = ' (1 + q 2 )

(10.8)

(10.9)

In the event of the meter design being such that vφc is not zero on purpose, an extension of the analysis to be presented in the following to account for this design

10.2 Physical and Mathematical Background

173

feature is straightforward. However, carrying out the analysis to cover this design feature also would only result in additional algebraic complexity without a corresponding enhancement in transparency, so a treatment in such generalised terms is not warranted at this stage. We will therefore restrict ourselves in the following to vφc = 0. For small values of αc which may be expected to be the case over the operating range of the turbine ﬂow meter, (10.9) may be written to show explicitly the relationship between the angle αc , the velocity proﬁle vxc , and the rotor angular velocity Ωc . This is: 1 1 ' − . (10.10) αc = sin αbl (1 + q 2 ) tan αbl The ﬂow rate Q is related to the velocity proﬁle uxc through R Q = 2π

Ro Ro ruxc dr = 2π rvxc dr = 2π r|vc | cos (αbl − αc ) dr , Ri

0

(10.11)

Ri

which may be approximated for small αc as Ro Q = 2π r|vc |(cos αbl + sin αbl αc )dr .

(10.12)

Ri

Under calibration conditions in question here, the expression for the aerodynamic torque on the rotor, (10.5), denoted as Taeroc , may be written as follows: Taeroc

Ro 2 2 = N ρ CL r(vxc + r2 ΩRc )Lbl αc cos αbl dr ,

(10.13)

Ri 2 2 + r2 ΩRc ). It it illuminative to rewrite Taeroc in a where vc2 has been written as (vxc form that brings out more explicitly its relation to the overall ﬂow rate through the device, Q, and the other relevant parameters associated with the ﬂow and the ﬂow ˆ meter geometry. Introducing for this purpose the notation Q(r) deﬁned through

r ˆ Q(r) =

r vxc dr ,

(10.14)

Ri

and the abbreviations with the dimensions of a length, viz. CL Lbl , tan αbl

(10.15)

CL Lbl cos αbl , tan αbl

(10.16)

Fmeter1 (r) = and Fmeter2 (r) =

174

10 Errors of Turbine Meters Due to Swirl

the expression (10.13), together with (10.10), may then be written as follows: Ro ' Taeroc ˆ = Fmeter1 vxc (1 + q 2 ) − Fmeter2 vxc (1 + q 2 ) Q Nρ Ri R o ' d ˆ Fmeter1 vxc (1 + q 2 ) − Fmeter2 vxc (1 + q 2 ) dr . − Q(r) dr Ri

(10.17) The expression for Taeroc in (10.13), which is complete within the framework of the physical hypotheses set out earlier, contains some physical features which are not easily visible at ﬁrst sight but are of signiﬁcance when the theory derived is to be employed to set its outcome against experimental observations of turbine ﬂow meter performance. We therefore prepare the ground for uncovering these features by deriving approximate forms that render the hidden structure more transparent. Rational approximations for Taeroc may be obtained by examining the limiting cases of (10.13) or (10.17) when the dimensionless quantity q(r) is either large or small. If q(r) assumes large values everywhere within the region Ri < r < Ro , it is permissible to replace (1 + q 2 ) by q 2 , on which Taeroc according to (10.13) or (10.17) gets reduced to the following form: Taeroc = ΩRc Nρ

Ro Ro 2 2 vxc r Fmeter1 dr − ΩRc r3 Fmeter2 dr . Ri

(10.18)

Ri

The corresponding expression for small values of q(r) is as follows: Taeroc = Nρ

Ro 2 vxc r(Fmeter1 − Fmeter2 )dr Ri 2 +ΩRc

Ro Fmeter1 3 − Fmeter2 dr . r 2

(10.19)

Ri

Two features worthy of special note in the expression for the driving (aerodynamic) torque, (10.13), or (10.17), which are retained in the approximations too are the following: • Taeroc taken as a whole is not proportional to the ﬂow rate Q. T aeroc may be regarded as a sum of two contributions, only one of which is proportional to the ˆ evaluated between the limits Ro and Ri is ﬂow rate Q. It may be noted that Q the ﬂow rate through the ﬂow meter, Q. The other contribution is a weighted integral of a function of the axial velocity proﬁle vxc where the weights further depend upon geometrical design details of the ﬂow meter geometry. • Taeroc is not expressible through Q alone, even when a nonlinear function of Q is permitted. Factors governed by meter geometry enter into the expression of Taeroc .

10.2 Physical and Mathematical Background

175

The latter property in particular, it will shortly be seen, has far reaching implications regarding casting the error of the turbine ﬂow meter due to swirl in a concise and universal form. To state the result in advance, this error will not be expressible in such a general form that is independent of ﬂow meter geometry and hence applicable to all ﬂow meters. Besides, in the approximations for neither of the limiting 2 contain vxc , the axial velocity cases, (10.18) and (10.19), does the coefﬁcient of ΩRc component determining the ﬂow rate Q, and it may therefore be conjectured that the exact expression retains this feature in the whole range of q too, although perhaps only approximately. Furthermore, for small q, the aerodynamic torque T aeroc does not contain any term linear in ΩRc . In contrast, for large q there is a term linear in ΩRc which is linear in vxc too. Although the aerodynamic torque is not proportional to the ﬂow rate, it is convenient for the purposes of comparison at the end of this section to write T aeroc formally as a product of Q and a quantity FT as follows: Taeroc = QFT . ρ

(10.20)

Written as above, the coefﬁcient FT with the dimensions of a length squared divided by time, besides containing a rest dependence on Q, is inﬂuenced by characteristics governed by the ﬂow meter design through Fmeter1 and Fmeter2 . The meaning of FT follows from a straightforward comparison of (10.20) with (10.17). A sample survey of published catalogue data on commercially availabe ﬂow meters indicates that they are designed to operate with the rotor angular velocity Ω R lying mostly in the range of the dimensionless parameter (RΩR )/(Uref ) between 0.9 and 1.5. Therefore, although the approximations (10.18, 10.19) do indeed preserve the salient feature of Taeroc not being proportional to Q, their applicability to evaluate meter characteristics theoretically is rather limited. The nature of dependence of the aerodynamic torque on the ﬂow rate is rather complex, in view of which the case for an examination of the aerodynamic torque obtainable from solving the full relevant equations of motions for the problem, say along lines set out in Chap. 11, is indeed strong. Such solutions, even though feasible for a few selected cases only, would, through a judicious comparison with suitable approximations, guide in assessing the regions of validity of the approximations, which would be useful for further analysis. For the present, the approximations provide essentially a qualitative insight into the phenomenon, leaving no option but to resort to (10.13) or (10.17) for quantitative evaluation. Summing up, given a certain ﬂow rate Q which sets the Reynolds number, there will be a certain proﬁle uxc , and hence vxc , from which the aerodynamic torque Taeroc follows from (10.10, 10.13). Taken together, (10.11-10.13) deﬁne the relationship between the ﬂow rate Q and the rotor angular velocity ΩRc under calibration conditions. This relationship is, of course, obtained experimentally during calibration. Therefore, the effects of the induced velocity ﬁeld due to the vortex system in the wake of the rotor when the oncoming ﬂow is the calibration ﬂow are accounted for.

176

10 Errors of Turbine Meters Due to Swirl

Measurement Conditions with Swirl in the Oncoming Flow The presence of swirl in the oncoming ﬂow implies that the azimuthal velocity component at the inlet section to the ﬂow meter (A-A in Fig. 10.1), u φ is non-zero. A crucial step in our study of the effects of swirl on the turbine ﬂow meter reading is characterization of the swirl in the oncoming ﬂow through a dimensionless parameter S . It is convenient, but not necessary, to regard S as, say, the value of the absolute maximum of the angular velocity of the ﬂuid in the laboratory coordinate system, referred to the angular velocity of the ﬂow meter rotor, i.e., uφ ω || = ±|| || , (10.21) S = ±|| ΩR rΩR with the sign + or − denoting whether the azimuthal velocity in the oncoming ﬂow is in the same direction as that of rotation of the ﬂow meter rotor or otherwise. A cursory survey through catalogues of commercially available turbine ﬂow meters indicates that they mostly operate at values of ΩR for which the parameter S assumes values in the range 0.025 − −0.25 in our experiments to be shortly described, which may therefore be regarded as small. In measurement, ﬂow conditions exist at the inlet section to the ﬂow meter that differ from those at calibration at the same ﬂow rate, and we denote the axial and azimuthal velocity distributions at this section (A-A in Fig. 10.1) as u x and uφ respectively. The radial distribution of this set of ﬂow quantities undergoes a change from the inlet section of the ﬂow meter to the inlet section to the blading, section A-A to B-B in Fig. 10.1, that depends upon meter geometry and hence differs from meter to meter. However, it is reasonable to hypothesize that the change is adequately described by Euler’s equations of motion. If furthermore, the change in ﬂow crosssectional area from A-A to B-B is also moderate, the parameter S is suited to characterize ﬂow perceived the blading also, and we may write the azimuthal velocity at the section B-B, vφ , in terms of the local angular velocity ω(r) as ω(r) vφ = = S fω (r) . rΩRc ΩRc

(10.22)

The axial velocity prevalent at the section B-B, vx (r) may also be written in terms of S as vx (r) = 1 + S fv (r) . (10.23) vxc (r) The condition that axial velocity distribution departs from that at calibration without any change in the ﬂow rate as such imposes an integral constraint on f v (r) which is as follows: Ro rvxc fv dr = Q1 = 0 . (10.24) Ri

The above constraint shows that fv (r) has to change sign at least once within the region Ri < r < Ro . The precise form of fv (r) depends upon the method of swirl generation in the experiment and of course on the design features of the ﬂow meter.

10.2 Physical and Mathematical Background

177

The change experienced by the rotor angular velocity, ∆Ω, as a consequence of the swirl may be written as S ΩR1 , or ΩR1 ΩR = 1 + S + O(2S ) . ΩRc ΩRc

(10.25)

Consequently, the azimuthal velocity at the inlet section to the blading (B-B in Fig. 10.1), vφ may also be expected to differ from zero, which in turn modiﬁes the velocity triangle. The modiﬁed velocity and force triangles are shown in Fig. 10.2b. Here again, as already set out, we work with the hypothesis that at any radial location r the velocity vector v and the force vector dFaero are perpendicular to each other. Under these conditions the angle of incidence α in (10.6) or (10.7) may be written as (10.26) α = αc + S α1 + O(2S ) , where αc is given by (10.10) and α1 is as follows: ΩR1 2 − fω − fv q ΩRc . α1 = − 3 (1 + q 2 ) 2 sin αbl

(10.27)

The functions of the radius r, these are q, fω and fv , are deﬁned through (10.8, 10.22) and (10.23) respectively. Writing the aerodynamic driving torque under measuring conditions, T aero given by (10.5), as an asymptotic expansion in terms of the small parameter S as follows, Taero = Taeroc + S Taero1 + O(2S ) ,

(10.28)

and expanding the integrand in (10.5) also in terms of S , shows that the expression for Taero1 is a sum of integrals. Taero1 may be written as Taero1

Ro = N ρ CL r(Tα + Tv + TΩ + Tω )Lbl αc cos αbl dr ,

(10.29)

Ri

where the functions in the integrand Tα , Tv , TΩ and Tω are as follows: 2 2 + r2 ΩRc ), Tα = α1 (vxc

(10.30)

2 , Tv = 2fv αc vxc

(10.31)

TΩ = 2ΩR1 αc r2 ΩRc ,

(10.32)

2 . Tω = −2fω αc r2 ΩRc

(10.33)

and The expression for Taero1 in (10.29) shows that to a linear approximation in S , the change in the aerodynamic driving torque experienced by the rotor is a weighted sum of the effects due to the change in the angle of incidence α1 , in the axial velocity

178

10 Errors of Turbine Meters Due to Swirl

proﬁle at the inlet to the blading fv , in the rotor angular velocity ΩR1 and in the swirl induced azimuthal velocity in the ﬂow fω . It may also be noted that α1 itself is a linear weighted sum involving ΩR1 , fv and fω , see (10.27). Taero1 is therefore a weighted sum of ΩR1 , fv and fω . The weights themselves, as an inspection of (10.30, 10.31, 10.32, 10.33) shows, are integrals of quantities that depend upon the details of the departure of the oncoming ﬂow from the calibration state and on the geometrical details of the ﬂow meter. The former depends upon the mode of swirl introduction into the ﬂow, whereas the latter is a design feature of the ﬂow meter. 10.2.2 Braking Torque on the Turbine Flow Meter Rotor In Baker (1993, 2000), the braking torque Tbr is itself written as a sum of four contributions, which are named as that due to bearing drag, hub disk friction drag, blade tip clearance drag and hub ﬂuid drag. For details of the characteristics of these contributions the reader is referred to Baker (2000) and original literature cited therein. The precise form of these individual contributions, and in particular the parameters affecting them and their dependence on these parameters, is still a matter of continuing investigations. For our present purposes, however, it sufﬁces to note that T br depends upon the rotor angular velocity ΩR . We also wish to add that the relationship between the contribuition to Tbr and ΩR is linear as long as the speciﬁc braking mechanism is adequately describable by classical lubrication theory involving the Stokes’ approximation, and bearing drag may be expected to follow this behavior. Under this hypothesis Tbr depends linearly upon ΩR and may be written as follows: Tbr = Bbr ΩR .

(10.34)

The coefﬁcient Bbr may then be identiﬁed as arising from bearing drag. It has the dimensions of a torque multiplied by time and is dependent upon geometrical particulars of the bearing and lubricant viscosity. The braking torque under measurement conditions, T br , may also be written as an asymptotic expansion in terms of the parameter S as follows: Tbr = Tbrc + S Tbr1 + O(2S ) ,

(10.35)

where Tbrc is the braking torque under calibration conditions and the dependence of each of the terms, Tbrc and Tbr1 , on ΩR is itself expressible in the form according to (10.34). The torque balance on the ﬂow meter rotor (10.1) then reduces to Tbrc = Taeroc ,

(10.36)

Tbr1 = Taero1 .

(10.37)

and The relation between the rotor angular velocity under calibration conditions Ω Rc and the ﬂow rate Q which is obtained experimentally during calibration, obeys (10.36), where Taeroc is given by (10.17), or (10.20), and Q by (10.12). This leads to the

10.3 Experiments and Data Reduction

179

following relation between the rotor angular velocity, ΩRc , and the ﬂow rate Q under calibration conditions: QFT (10.38) ΩRc = Bbr The relation between the K factor commonly quoted for linear ﬂow meters - see e.g., Baker (2000) - and the quantities FT and Bbr introduced in this work is evident on inspection. We wish to redraw the reader’s attention to the point made earlier, viz. that FT cannot be strictly a constant, even for a speciﬁc ﬂow meter. The corresponding relation between the departure from the calibration state and the rotor angular velocity under measuring conditions are contained in (10.29, 10.37). We postpone employing this relation to get an estimate of the error in metering to the next section in this chapter and note the following for the present: Given proﬁles of the axial and azimuthal velocity components, u x and uφ entering the turbine ﬂow meter at section A-A (Fig. 10.1) under measurement conditions, the components vx and vφ at the blade inlet section may be calculated if geometrical design details of the turbine ﬂow meter are known, and which, in turn, leads to Ω R1 and Q1 , which are the quantities sought.

10.3 Experiments and Data Reduction Experiments were conducted in a specially designed facility in which the turbine ﬂow meter could be exposed to an oncoming ﬂow with swirl. Swirl over a wide range of the parameter S could be generated in this facility by a rotating tube bundle, identical in type and dimensions with that described in Rocklage-Marliani et al. (2003). The direction of rotation of the swirl generator could also be reversed, so that both positive and negative values of the parameter S could be realised. Provisions were made in this facility for an independent measurement of the ﬂow rate. The latter was regarded as the reference measurement for the present purpose, denoted Q ref . For a description of this facility the reader is referred to Chaps. 4, 8. The uncertainty of the ﬂow rate in the reference measurement is ±0.3%. The ﬂow rate “read” by the turbine ﬂow meter, Qmeas , together with the reference measurement thus enabled a plot of the error “read” by the turbine ﬂow meter vs. the swirl parameter S to be obtained experimentally. A schematic diagram of the facility is shown in Fig. 10.3. It is essentially a pipe 100 mm inside diameter and 120 diameters long, at the downstream end of which the turbine ﬂow meter is installed. In this facility, the swirl generator could be mounted at a certain number of chosen locations along the pipe upstream of the ﬂow meter. The evolution characteristics of both the mean and ﬂuctuaing motion in the ﬂow generated by this device over the range of ﬂow parameters (Reynolds number and swirl number) of interest in the present work were ascertained through extensive experiments conducted in a separate facility, see Rocklage-Marliani et al. (2003). The turbine ﬂow meter could thus be exposed to an oncoming ﬂow with swirl whose distribution characteristics along the radius at the ﬂow meter inlet section were known.

180

10 Errors of Turbine Meters Due to Swirl

For purposes of later reference in Sect. 10.4 we wish to note here a salient property of the distribution characteristics of the azimuthal velocity proﬁles at different streamwise locations within the measurement regime. It is that the azimuthal velocity proﬁles essentially scale with the angular velocity of the swirl generator, ω S , i.e., they can be written as (uφ )/(rωS ) = fφ (x/R, r/R). In the core region with a solidbody like rotation the function fφ is independent of r, however slowly changing with respect to x.

Fig. 10.3. Schematic diagram of experimental facility.

This facility was operated with the ﬂow rate set at a certain level and the angular velocity of the swirl generator varied over a wide range of parameters. The ﬂow rate “read” by the turbine ﬂow meter, Qmeas , is the value read off from the calibration curve of the ﬂow meter corresponding to the angular velocity of the ﬂow meter rotor. The reference measurement of the ﬂow rate, Qref , which is done independently, then enabled the error to be evaluated. In order to set the experimental outcome against the theory meaningfully, the error evaluation from the measurement data should correspond in spirit to the error deﬁnition in the theory. In the present context, satisfying this requirement, which is generally considered too obvious to deserve any special reference to, is not as straightforward as it may seem to be at a ﬁrst glance. The reason is that the theory deals only with such disturbances traceable to swirl in the oncoming ﬂow, whereas in the experiment the ﬂow meter “perceives” all the disturbances that are present regardless of their origin. The scope of the theory is clear on examining the mathematical formulation in which is the disturbance in both the axial and azimuthal velocity components is characterized through one and the same parameter S . In the experiment on the other hand, the departure from the calibration state “perceived” by the ﬂow meter, depends upon the method of disturbance generation. The question of parametrical characterization of the disturbance in the experiment has therefore to be addressed. In particular, characterization through a single parameter may not be feasible. The disturbance caused by a stationary honeycomb as in the experiments reported in Chap. 4 or when the tube-bundle of our present is not set in rotation illus-

10.3 Experiments and Data Reduction

181

trate this point. In short, it is quite possible to create a departure from the calibration state in the axial velocity component only without a inducing a corresponding azimuthal component, a case clearly not encompassed by the present theory. In order to set the present experimental outcome against theory, it is therefore necessary to evaluate the measurement data according to a procedure that separates out the error traceable to swirl. In summary, for a proper comparison between experiment and theory, therefore, the following two points need to be addressed: • Deﬁnition of error. • Data reduction procedure for extraction of the error due to swirl. 10.3.1 Deﬁnition of Error The theory presented earlier relates changes in the ﬂow rate and in the angular velocity of the ﬂow meter rotor to departures of the velocity proﬁles in the oncoming ﬂow from their calibration state. Within this framework, two distinct approaches of ascertaining the error can be identiﬁed. • The ﬁrst is the change in angular velocity experienced by the ﬂow meter rotor at a certain ﬂow rate. This corresponds to keeping track of ΩR1 under the integral constraint (10.24). • The second is evaluating the change in the ﬂow rate at ΩR1 = 0. This corresponds to evaluating the integral on the left hand side of (10.24). Deﬁning the error as (Qmeas − Qref ) corresponds to an experimental measurement of this integral. The second approach is the simpler of the two from the point of view of conducting the experiment in our setup in which there was no provision made for a direct and independent measurement of the rotor angular velocity. 10.3.2 Extraction of Error Due to Swirl The quantity (Qmeas − Qref ) contains, as stated earlier, the total error due to the disturbance actually present in the experiment. In particular it includes the error caused by the departure of the axial velocity proﬁle from the calibration ﬂow in the experimental conﬁguration in question. Strictly speaking, the departure in this quantity caused by a rotating tube bundle at any particular streamwise location would depend upon the angular velocity of the tube-bundle, and need not necessarily be the same as with the tube bundle stationary. However, the experiments of Rocklage-Marliani et al. (2003) indicate that, in the ﬂow in our experiment the Reynolds-averaged axial velocity proﬁle is relatively insensitive to tube-bundle rotation. The main source of error for the ﬂow meter reading is the azimuthal velocity proﬁle. Therefore an estimate for the contribution to the error due to swirl may be obtained by subtracting out the error measured when the swirl generator is in its place but with the tube-bundle not set in rotation. We denote this quantity by the subscript 0, i.e., as (Q meas −Qref )0 .

182

10 Errors of Turbine Meters Due to Swirl

A simple procedure for extracting the error caused by swirl is then forming the difference (Qmeas − Qref ) - (Qmeas − Qref )0 . The error Q evaluated as Q =

(Qmeas − Qref ) − (Qmeas − Qref )0 Qref

(10.39)

has been plotted against the rotation number, Ro, in Fig. 10.4. The straight line drawn is one that satisﬁes the least square error criterion through the measured data. 0.06 0.04

HQ

0.02 0 -1.2

-0.8

-0.4

-0.02 0

0.4

0.8

1.2

-0.04 -0.06 Ro

a) x/D=4

0.06 0.04

HQ

0.02 0 -1.2

-0.8

-0.4

-0.02 0

0.4

-0.04 -0.06 Ro

b) x/D=14

Ŷ Re=50,0000.5% Ÿ Re=200,0000.5% Fig. 10.4. Error in ﬂow metering.

0.8

1.2

10.5 Conclusions

183

10.4 Discussion The results of the analysis in Sect. 10.2 show that the output signal of the turbine ﬂow meter, and hence the error in reading, depends not only upon details of the radial distribution of the disturbance but also on geometrical design features peculiar to a particular turbine ﬂow meter, viz. Fmeter1 and Fmeter2 . These quantities also enter the problem not by themselves but as products integrated across the blade height. Furthermore, the design data are rarely, if ever, available in sufﬁcient detail to the user, which makes it a virtually hopeless undertaking to try to formulate error characteristics of turbine ﬂow meters in universal terms. Against the background of this limitation, the task of comparing the error measured in the experiment with the theory presented can be charted out to proceed along several conceptually different routes. The most immediately obvious choice would be to set the measured quantity (Qmeas − Qref ) against the corresponding quantity evaluated from theory. Such a conceptually simple and straightforward route of comparison, although meriting consideration for the study a particular design as such for which the relevant design data are available, does not entirely fall within the scope of the present work. The reason is that such a study of a single ﬂow meter diverts attention away from the main objective set for the present work, which is to illuminate the gross characteristics of the behavior of swirl-induced error in ﬂow-rate measurement through turbine ﬂow meters. This purpose is better fulﬁlled by the alternative route of examining the “scaling characteristics” of the error, on which we focus our attention in the following. For this purpose it is only necessary to note an outstanding point made by the theory that the relation between the error and the strength of the disturbance of a given shape in the oncoming ﬂow is a linear one. The strength of the disturbance used in the theory is characterized by S , and a suitable measure for the disturbance in the experiment which, incidentally is primarily in the azimuthal velocity, is the rotation number Ro. The experiments of Rocklage-Marliani et al. (2003) indicate that at any particular location downstream of the swirl generator within the region x/D < 14, the azimuthal velocity distribution scales with the rotation number Ro. The shape of the azimuthal velocity distribution itself is however dependent on x/D. For this reason, the linear relationship between the error Q and the swirl parameter S may be expected to be carried over to the pair Q and Ro, and the plot in Fig. 10.4 shows this indeed to be the case. The magnitude of the error itself at any particular location is primarily dependent upon the radial distribution of the azimuthal velocity at this location, and the differences in the slope of the Q vs. Ro plot are a manifestation of this property of the error. The plot also brings out the feature expectable on physical grounds, viz. the change in sign of error with respect to the direction of the swirl in the oncoming ﬂow.

10.5 Conclusions The main conclusion that can be drawn from the present study regarding the behavior of swirl-induced error in metering by a turbine ﬂow meter is as follows: The error

184

10 Errors of Turbine Meters Due to Swirl

shows a linear relationship with respect to the parameter characterizing the strength of the disturbance in the oncoming ﬂow, but the actual magnitude of the error depends both upon the radial distribution of the disturbance in the oncoming ﬂow and upon geometrical design features of the ﬂow meter.

References Baker, RC (2000) Flow Measurement Handbook. Cambridge University Press, Cambridge UK Baker, RC (1991) Turbine and related ﬂowmeters: I. Industrial practice. Flow Meas Instrum 2:147–162 Baker, RC (1993) Turbine ﬂowmeters: II. Theoretical and experimental published information. Flow Meas Instrum 4:123–144 Batchelor GK (1967) An Introduction to Fluid Dynamics. Cambridge University Press, Cambridge UK Leishman JG (2000) Principle of Helicopter Aerodynamics. Cambridge University Press, Cambridge UK Marshall JS (2001) Inviscid Incompressible Flow. John Wiley, New York Rocklage-Marliani G, Schmidts M, Vasanta Ram VI (2003) Three-dimensional laser-Doppler velocimeter measurements in swirling turbulent pipe ﬂow. Flow, Turbulence and Combustion 70:43–67 Schlichting H, Gersten K (2000) Boundary Layer Theory. Springer, Heidelberg Schlichting H, Truckenbrodt E (1962) Aerodynamik des Flugzeuges, Erster Band. Springer, Heidelberg Schmidts M, Marliani G, Vasanta Ram VI (1998) A study of the metering error of turbine ﬂow meters caused by swirl in the ﬂow. FLOMEKO 98, Proceedings:399–403 Van Dyke M (1975) Perturbation Methods in Fluid Mechanics, Annotated edition. The Parabolic Press, Stanford California USA

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter Ernst von Lavante Institut für Strömungsmaschinen, Universität Essen, 45117 Essen, Germany

Three-dimensional internal ﬂow in a complete stage of a modern turbine ﬂow meter was numerically simulated in order to investigate the ﬂow ﬁeld in general, and possibilities of improving the accuracy of the meter in particular. The governing equations of the three-dimensional, compressible, viscous ideal gas ﬂow were solved numerically using two simulation programs, the commercially available Fluent and the academic code ACHIEVE. The resulting ﬂow ﬁelds, including the details of various ﬂow effects, are discussed. By close consideration of the ﬂow ﬁelds, the so-called pressure shift can be explained as an effect depending on the Reynolds number.

11.1 Introduction The requirements imposed on the measuring systems used in global natural gas market are very high. One ﬂow meter that fulﬁlls the high requirements and is capable of measuring very accurately even under high pressures in ﬁeld use is the turbine ﬂow meter. In the past, most of the investigations of turbine ﬂow meters were done analytically and experimentally. Several authors focused on the development of an equation of motion of the meter by relating the driving torques of the ﬂuid to the friction torques due to friction of the ﬂuid on the hub, in the tip gap or in the bearing (Baker, 1993; Lehmann, 1990; Lee and Karlby, 1960). Many investigations (Brümmer, 1997; Lehmann, 1990; Mickan et al., 1996) studied the effect of pulsations or proﬁle deformations on the accuracy of the meter. In the last two decades research was done on self-correcting and self-checking gas turbine meters. One working principle is based on the coupling of two rotors. A sensor rotor small distance downstream from the main rotor senses and responds to changes in the exit angle of the ﬂuid leaving the main rotor (Lee et al., 1982; Lee and Karlby, 1960). Another working principle assumes two decoupled rotors, where the second, the so-called reference rotor, works independently of the ﬁrst stage. The resulting two-

186

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

Master Meter Stator (1)

Ref. Stator (2)

Rotor (1) Fig. 11.1. Cutaway Drawing of AccuTest

Ref. Rotor (2) TM

Turbine Flow Meter.

stage turbine meter AccuTestTM (Fig. 11.1) has been developed and investigated in recent years experimentally and analytically by Schieber (1998, 2000). In this work, the ﬂow ﬁeld in an 6" AccuTestTM turbine meter has been investigated numerically. To this end, the ﬂow in the two-dimensional blade-to-blade plane consisting of both stages has been simulated ﬁrst, proving the feasibility of the present study. Consequently, the unsteady three-dimensional ﬂow ﬁeld has been investigated in the ﬁrst stage of the meter, aimed to describe the main ﬂow effects occuring in the meter and to ﬁnd promising design modiﬁcations to increase the accuracy of the meter. This chapter will concentrate on the three-dimensional results, aiming at a detailed discussion of the ﬂow features in the ﬂow ﬁeld typical of modern turbine ﬂow meters.

11.2 Theory of Operation The metering principle of the turbine ﬂow meter is based on a volumetric measurement. Here, parts of the ﬂuid with a certain deﬁned volume given by the geometry of the meter are integrated in time and the transported amount of ﬂuid is displayed (Bonﬁg, 1997). The turbine meter is an inferential ﬂow meter meaning that a certain amount of ﬂuid is not directly measured but related to the frequency of a rotor that is impelled by the ﬂow through the meter. The frequency f of the rotor is measured and related to the volume ﬂow rate by the deﬁnition of the so-called k factor. The k factor is deﬁned as the ratio of the total number of output pulses to the actual volume of gas that passes through the meter. When considering same time periods, k becomes the

11.2 Theory of Operation

187

˙ ratio of the output frequency over the actual ﬂow rate Q: k=

f Q˙

(11.1)

The k factor is a function of the operating condition (pressure, temperature and density of the ﬂuid) and the physical dimensions of the meter. Its value is determined in calibration tests. Figure 11.1 shows a cutaway drawing of the AccuTestTM turbine ﬂow meter that was investigated in this project. The design of the ﬂow meter consists of two parts. The front part, with the gas entering from the left, is a slight modiﬁcation of an existing, conventional meter. The second part is a so-called master meter, designed to achieve the self-proving feature of the meter. It is a separate free-running turbine meter cartridge; its blade wheel rotates in opposite direction to the ﬁrst rotor. Rotor speed is measured with an electronic sensor that detects the rotor blade tips as they pass the proximity of the sensor. The function of the master meter is to evaluate the accuracy of the main meter. Schieber (1998) derived a working equation of the two-stage turbine meter. He deﬁnes the Accuracy of the meter Acc as the ratio of the measured volume ﬂow rate Q˙ and the actual volume ﬂow rate Q˙ norm . Under the assumptions that the accuracy of the reference rotor is equal one, the bulk ﬂow is one-dimensional, inviscid, incompressible, and steady in time, and the volume ﬂow rate at the main rotor equals the volume ﬂow rate at the reference rotor, one obtains for the working equation: AccSPM =

fmain kref Acccal fref kmain

(11.2)

where SPM denotes the self proving meter and Acccal ist the accuracy determined during the calibration. The subscripts main and ref describe the ﬁrst and second stage of the turbine meter, respectively. In order to make 11.2 valid, it is assumed that the volume ﬂow rate inferred from the reference rotor is the actual ﬂow rate, the rotors are completely ﬂuiddynamically decoupled, factors inﬂuencing the performance of the upstream main rotor will not inﬂuence the downstram reference rotor and the master meter performance will not change from initial calibration after extended use in the ﬁeld. One major beneﬁt of the two-stage turbine meter over a conventional turbine meter is a reduced need for inspection and maintenance, requiring only one person with a laptop or a so-called electronic ﬂow computer. Furthermore, the meter performance is monitored for immediate detection of accuracy changes and the effects of ﬂow disturbances. Installation effects are weaker due to an improved guide vane design of the ﬁrst stage. In order to optimize the design and accuracy of the two-stage turbine meter, the ﬂow ﬁeld inside the meter was analyzed. Since it is rather difﬁcult to investigate the ﬂow in the meter experimentally, it was decided to simulate the ﬂow numerically using two computer programs for comparison. These two programs are described brieﬂy in the following part 3 of this chapter.

188

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

11.3 Numerical Tools: Fluent and ACHIEVE Nowadays, several computer programs to simulate the 3-dimensional, viscous, unsteady and turbulent ﬂows by solving the Navier–Stokes equations are available. Not all of them are capable of handling unsteady stator/rotor interactions correctly, however. In the present work, the commercial solver Fluent (1998) has been chosen due to its advertized capability to handle the case under investigation. One problem that most commercial solvers have is a high amount of numerical dissipation, which is implemented in order to enhence the robustness of the code. This implies that since the effective Reynolds number could be up to one or more orders of magnitude lower than in the reality, some viscous ﬂow effects could not be modeled accurately. For this reason, the more accurate code ACHIEVE (Algorithm for Solving Chemically Reacting Viscous Flow Problems) (Kallenberg, 1999; Perpeet, 2000; Yao, 1997), also capable of calculating the ﬂow ﬁeld in the given turbine meter, has been used at the same time. ACHIEVE has been developed at several academic institutions around the world under the guidance of the author, and has, due to its formulation, inherently smaller numerical dissipation than Fluent. Both codes are compared in the following Table 11.1. Table 11.1. Features of the two solvers. ACHIEVE

Fluent

Roe’s ﬂux difference splitting in ﬁnite volume form two-stage modiﬁed Runge–Kutta time stepping k-ω-turbulence model compressible ﬂow ideal gas law structured grid

segregated/coupled implict solver second order implicit method in pseudo-time k--turbulence model compressible ﬂow ideal gas law mixed grid(structured in b.l., unstructured in far-ﬁeld) multiblock structure; computation on a single CPU

multiblock structure; parallel computation on a PC cluster

11.4 Three-Dimensional Simulations: Geometry and Setup In the ﬁrst part of this section, the 3-dimensional model including the corresponding computational grid of the turbine ﬂow meter will be explained. In the next section, the results of the simulation including the stator/rotor interaction, ﬂow in the tip gap and the inﬂuence of boundary layers on the angel of incidence β i will be discussed. In this work, the inﬂuence of boundary layer effects and the ﬂow in the tip gap were to be investigated, making a high grid resolution close to the walls essential. In

11.4 Three-Dimensional Simulations: Geometry and Setup

189

order to achieve a reasonable number of cells in the computational grid it was decided to simulate the ﬂow in the ﬁrst stage only. Since the ﬁrst stage closely resembles the convential turbine ﬂow meters, the current analysis of the present results should be of general beneﬁt. Besides, the geometry of the ﬁrst stage is more complicated than that of the second stage because of the central body (hub) in the inlet section of the guide vanes of the ﬁrst stage.

Fig. 11.2. Isometric view of the ﬁrst stage of the turbine ﬂow meter; left: front view looking downstream; right: view from the back looking upstream.

Figure 11.2 shows an isometric view of the geometry of the ﬁrst stage of the turbine meter, as seen from the front (inlet) and from the back (outﬂow), assumed in the present simulations and plotted without the outer casing of the meter. The grid had approximately 1, 000, 000 cells, covering all relevant boundary layers at hub, casing and the blades. The ﬂow enters the meter axially in the positive x-direction (see Fig. 11.2). It passes at ﬁrst the central body (hub) with guide vanes of the ﬁrst stator. Further downstream it ﬂows through the rotor. In the present simulation, the speed of rotation of the rotor is a predetermined input, being initially speciﬁed. It was obtained from corresponding experimental data. Only one blade passage of the guide vanes and the rotor have been simulated to save computational time. Therefore, the boundaries between each passage had to be treated as periodic and the incoming velocity proﬁle was assumed to be symmetric fully developed turbulent ﬂow. In the actual meter, there are 16 blades in both the rotor and the stator, making the assumption of periodicity possible. In the display of the results, the computed ﬂow ﬁeld has been projected to the remaining 15 passages. Figure 11.3 shows the structured grid in the one-passage model conﬁguration. In circumferential direction 58 grid points resolved the blade to blade plane. An exponential distribution close to the wall provides very small distances of the ﬁrst cells to the wall so that at least 10 points normal to the wall were within the boundary layer. In radial direction 51 grid points were distributed over the height of the blade with exponential distributions close to the walls. 15 grid points were equally spaced in the 0.8 mm heigh tip gap.

190

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

Fig. 11.3. Structured grid of the model conﬁguration.

The three-dimensional Fluent simulations were performed on a 800 MHz LinuxPC with a 768 MB RAM disk. The simulation of one operating point until a time periodic state is reached, took about 3–4 weeks to perform. In addition to the excessive CPU time, the static pressure at the outﬂow boundary had to be adjusted iteratively in order to obtain the required mass ﬂow.

11.5 Discussion of Results Four different operating points of the meter were simulated to study the 3dimensional ﬂow effects. The inﬂuence of different ﬂow capacities on the ﬂow ﬁeld in the meter was investigated by simulations at 10%, 50% and 100% of the maximum ﬂow capacity Q˙ max at an operating pressure of 1 bar. The inﬂuence of varying pressure was taken into consideration by the simulation at 10% of maximum capacity and a pressure of 10 bar. The ﬂow at 1 bar and 100% capacity corresponds to the approximately same Reynolds number as the simulation at 10 bar and 10% of maximum capacity. The Reynolds number, calculated based on the incoming velocity of the meter and the diameter of the turbine meter is Re = 150, 000. Due to the fact that the contour plots at all investigated operating points look very similar, the ﬁgures in this chapter concentrate on the simulation at p = 1 bar and Q˙ = Q˙ max . The global range of the computed velocities in the entire ﬁrst stage was between 12 m/s and 42 m/s. As the ﬂow enters the guide vanes the cross-sectional area is reduced by the central body, accelerating the ﬂow as indicated in Fig. 11.4 by the decrease of pressure along the central body. Here, the static pressure on the walls of the entire stage is shown. A stagnation point develops at the front of the central body. A video animation of the time-dependent static pressure on the walls revealed that there is an inﬂuence

11.5 Discussion of Results

191

Fig. 11.4. Static pressure contours on the walls of the entire stage.

of the rotor-stator interaction on the ﬂow in the guide vanes all the way upstream to the stagnation point. The pressure at the stagnation point changed periodically with the blade-passing frequency. An examination of the ﬂow around the central body showed that no ﬂow separation occured at this part of the turbine ﬂow meter. As seen in Fig. 11.4 as a dark line on the rotor leading edge, a stagnation point develops at this location. Close to the hub and close to the tip the stagnation pressure decreases due to the reduction of the total pressure. This is due to the inﬂuence of the boundary layer that causes lower velocities close to the wall, resulting in smaller values of stagnation pressure. Immediately downstream of the leading edge the ﬂow accelerates, giving rise to a high risk of ﬂow separation in the subsequent deceleration zone. A region of relatively low pressure develops on the suction side close to the trailing edge indicating a risk of ﬂow separation as would be typical for turbine ﬂow. In Fig. 11.5 the static pressure on the rotor versus the non-dimensionalized length of the rotor is shown in three different radial planes for two time instants. The planes under consideration are described by the percentage height h of the rotor, where zero height is at the hub. The left graph in Fig. 11.5 shows the static pressure at the time t = 0 T 1 when the rotor leading edge is directly behind a guide vane. The pressure at the stagnation point (x/L = 0.01) reaches the greatest values in the 50% h-plane. Here, the boundary layer inﬂuences the pressure at the stagnation point since the velocity in the boundary layer approaches zero at the wall. In the 5% h-plane, pressure on the suction side close to the leading edge is greater than on the pressure side. This unfavorable effect is caused by the wake of the stator that reduces the axial velocity thus increasing the incidence angle. At the trailing edge, the pressure in the 50% h-plane reaches again the greatest values. The pressure distribution in the same planes of constant radius for the time when the leading edge of the rotor is in the middle between two guide vanes of the stator, t = 1/2 T1, is shown in the right graph in Fig. 11.5. Similar ﬂow effects as at the

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

103000

103000

102000

102000

101000

101000

Pressure [Pa]

Pressure [Pa]

192

100000

99000

98000

99000

98000 5% h 50% h 95% h

97000

t = 0 T1 96000

100000

0

0.25

0.5

x/L [-]

0.75

5% h 50% h 95% h

97000

t = 1/2 T1 1

96000

0

0.25

0.5

0.75

1

x/L [-]

Fig. 11.5. Static pressure on the rotor in different radial planes for two rotor positions relative to the stator.

time t = 0 T1 can be found for this rotor position. At t = 1/2 T1, the wake of the stator does not inﬂuence the ﬂow at the leading edge of the rotor so that the incoming ﬂow to the rotor is more uniform. The incidence angle does not change signiﬁcantly between the hub and the tip. The vortex formation and pressure downstream of the trailing edge vary in radial direction. In the plane close to the tip (h = 95%) of the blade no vortex develops at the top side of the ﬂap. In addition to the boundary layer effects, the inﬂuence of the tip gap affects the ﬂow at this location. A quantitative analysis of the inﬂuence of the boundary layer at the hub and the casing on the inlet and outlet angles in the rotor was acomplished by plotting the mass averaged angle (β1 − βs ) at the inlet and (β2 − βs ) at the outlet of the rotor versus the non-dimensionalized height of the rotor for the two times t = 0 T 1 and t = 1/2 T1 considered here. For both times the angle (β1,2 − βs ) is in the range between 3◦ –5◦ for a blade height of 30%–70%. Close to hub, h < 30%, and casing, h > 80%, (β 1,2 − βs ) increases up to values of approximately 14◦ . The inﬂuence of the boundary layer at the hub is stronger than at the casing because the tip gap reduces the above effects. In the design of the AccuTestTM turbine ﬂow meter, the radial distribution of the stagger angle βs is based on the Euler theory. The stagger angle βs is proportional to the rotational speed, being a function of radius and angular velocity. No friction effects were taken into consideration. In reality, friction inﬂuences the ﬂow. This phenomenon causes the variation of the reduction of the incidence angle in the presence of the boundary layer, shown in Fig. 11.6 (right). The axial velocity in the boundary layer is smaller than in the case without friction. Keeping the rotational speed constant for both axial velocities yields an increased angle of incidence in the boundary layer. No signiﬁcant inﬂuence of the wake from the stator on (β1,2 − βs ) can be seen at t = 0 T1 due to the mass averaging. It is easy to realize, however, that smaller

11.5 Discussion of Results

193

15

β1,2 − βs[deg.]

10

5

inflow, t = 0 T1 inflow, t = 1/2 T1 outflow, t = 0 T1 outflow, t = 1/2 T1

0

0

20

40

60

80

100

Nondimensionalized Blade Height [-]

Fig. 11.6. Inﬂow and outﬂow angles for two different rotor positions.

axial velocities in the wake of the stator at t = 0 T1 would increase the local angle (β1,2 − βs ). The effect of the radially varying incidence angles could be reduced by design of the rotor, but for the given application, this procedure would not be practical. Concentrating now on the ﬂow in the tip gap, Fig. 11.7 should be consulted, showing density contours and streamlines of absolute velocity in three axial planes (x = −0.01, x = 0.00 and x = 0.01) of the meter.

Fig. 11.7. Density contours and streamlines of absolute velocity in different axial planes of the ﬁrst rotor at two different positions relative to the stator. (a) behind the stator, (b) in the middle of the channel between two stator vanes.

194

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

In the plane close to the leading edge (x = −0.01), a vortex rotating counterclockwise originates in the tip gap due to the superimposition of the incoming ﬂow and the ﬂow in circumferential direction imposed by the rotation of the blade. Further downstream, in the middle of the rotor (x = 0.00), the pressure difference between the pressure and the suction side of the rotor becomes dominant as the driving force in the blade. Consequently, the streamlines indicate ﬂow from the pressure side through the tip gap to the suction side of the rotor. This behavior does not change further downstream. Next, the evolution of the axial velocity in the turbine ﬂow meter will be discussed in more detail. In order to obtain an overview of the axial evolution of the ﬂow in the turbine meter, the axial velocity proﬁles in selected axial planes are displayed in Fig. 11.8. The contours range from a velocity of 0 m/s to 42 m/s.

Fig. 11.8. Velocity proﬁles in selected axial planes.

At the inlet of the turbine meter, a fully developed velocity proﬁle (Re = 150, 000) can be seen. At the casing, there is a velocity minimum, indicating that a boundary layer exists at this location. Further downstream, the ﬂow has been accelerated due to the displacement effect of the central body. At the end of the guide vanes every velocity proﬁle in one blade section looks like a developed turbulent proﬁle in a rectangular channel ﬂow with boundary layers at the solid surfaces. Downstream of the guide vanes, the wakes behind the guide vanes are being mixed out by natural dissipation, making velocity distribution more uniform.

11.5 Discussion of Results

195

35

30

ucross / uax [%]

25

20

15

10

5

0 -0.15

-0.1

-0.05

0

Axial Coordinate x [m]

Fig. 11.9. Radial velocity component in the turbine stator.

Just upstream of the rotor, the reduction of the axial velocity in the wakes is smaller but still noticeable. They cause the coefﬁcient of moment c m of the ﬁrst rotor to vary in time. At the outﬂow from the rotor, the ﬂow has again been accelerated due to a reduction of the cross-sectional area. A wake forms behind the trailing edge of the rotor. While the ﬂow is convected downstream, the velocity proﬁles rotate in negative ω-direction because the deﬂection of the ﬂow in the rotor generates circumferential velocity components. At the outlet of the present conﬁguration, the velocity proﬁle has already become more uniform due to the mixing taking place. In order to analyze the described ﬂow phenomena quantitatively, Fig. 11.9 shows the radial velocity component, non-dimensionalized by the axial velocity, versus the axial coordinate in the stator. At the inlet of the meter no radial velocity component can be detected since the ﬂow was assumed to be undisturbed. When the ﬂow passes the central body of the ﬁrst stator, the ﬂow in the center of the meter follows the geometry of the central body. Therefore, the radial ﬂow component was originated and reached up to 32% of the axial velocity. Downstream of x = −0.14 m, the radial ﬂow components are reduced by the shape of the guid vanes, so that just at the end of the guide vanes (x = −0.03 m), no signiﬁcant radial velocity component can be detected. At x = −0.05 m, a small radial velocity component is generated by the wake of the stators, but is dissipated later within a short distance. The wake ﬂow of the stator is analyzed in more detail by plotting the axial velocity in the midplane of the meter versus the circumferential direction of one blade section in 5 axial planes that represent the region between the stator and the rotor. In all planes the inﬂuence of the wake is recognized as a velocity proﬁle deformation. In the plane immediately behind the guide vanes at x = −0.06 m the minimum velocity in the wake is only 9 m/s whereas maximum velocities of up to 43.5 m/s are reached. Closer to the rotor the minimum velocity in the wake increases due to

196

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

50 45

Axial Velocity u [m/s]

40 35 30 25 20 x=-0.02 x=-0.03 x=-0.04 x=-0.05 x=-0.06

15 10 5 0.062

0.063

0.064

0.065

0.066

0.067

z [m]

Fig. 11.10. Axial velocity (midplane) in the wake of the stator.

mixing up to 31 m/s. This non-uniform velocity proﬁle is the inlet proﬁle of the rotor and yields a variation of the coefﬁcient of moment in the rotor with time.

11.6 The Pressure Shift One of the main reasons for the present investigation was the explanation of the so-called pressure shift in the accuracy of the turbine ﬂow meters. This expression denotes an unfavorable change of accuracy with increased pressure of the metered gas.Suspecting that the cause of this effect is a change of the ﬂow in the rotor, it was decided to study the deﬂection angle as a function of varying ﬂow rate and pressure. Simulations of the ﬂow were performed between 10% and 90% of the maximum ﬂow capacity at low pressure (1 bar) and high pressure(10 bar). The angle of incidence was chosen to remain constant at βi = 2.5◦ . It should be obvious that the deﬂection angle ∆β = β2 − β1 must be a nonlinear function of the ﬂow rate Q˙ as long as the ﬂow rate is proportional to the axial velocity. For increasing ﬂow capacities Q˙ at an operating pressure of 1 bar the deﬂection angle ∆β decreases until approximately half the maximum ﬂow capacity was reached. For higher ﬂow rates it remained approximately constant. At a pressure of 10 bar, the deﬂection angle ∆β was approximately constant for every investigated ﬂow capacity. Plotting the deﬂection angle as a function of the Reynolds number (Fig. 11.11), a continuous line of a dependence of the deﬂection angle on the Reynolds number can be found. This is because the maximum ﬂow capacity at 1 bar corresponds to the same Reynolds number of the minimum ﬂow capacity at 10 bar, and the deﬂection angles are in both cases approximately 5◦ . A critical Reynolds number Recrit , describing the smallest Reynolds number for which the deﬂection angle ∆β remains constant can be given as approximately Recrit ≈ 100, 000.

11.6 The Pressure Shift

197

14.0

12.0

∆β [deg]

10.0

8.0

6.0

4.0

2.0

0.0

1 bar 10 bar 0

200000

400000

600000

800000

Reynolds number [-]

Fig. 11.11. Deﬂection angle ∆β of the ﬂow through the rotor versus Reynolds number.

The coefﬁcient of moment cm of the rotor is deﬁned as: cm =

Mpr + Mvisc ρref c2ref Aref

(11.3)

1 2

where M denotes moments due to viscous (visc) and pressure (pr) forces. The subscript ref denotes a reference state deﬁned as the inﬂow conditions. Aref is the area of the rotor blade calculated by the product of chord length and height of the blade. Figure 11.12 shows the dependency of the coefﬁcient cm on the Reynolds number, displaying a behavior similar to the deﬂection angle in two dimensions. c m ﬁrst

0.04

1 bar, 10 %

cm [-]

0.03

0.02

0.01

1 bar, 50 % 1 bar, 100 % 10 bar, 90 %

10 bar, 10 %

0.00 0

a)

500000

1000000

Re [-]

Fig. 11.12. Coefﬁcient of moment from three-dimensional simulation.

198

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

101

Mechanical Accuracy [%]

100.5

100

99.5

99

98.5

98

27 psi, old 100 psi, old 0

200000

400000

600000

800000

Reynolds Number [-]

Fig. 11.13. Mechanical Accuracy Acc versus Reynolds number.

decreases for smaller Reynolds numbers, then seems to reach a constant value for Reynolds numbers above a certain critical value. The coefﬁcient of moment determined at 100% ﬂow capacity and 1 bar pressure is very close to that at approximately the same Reynolds number for 10% ﬂow capacity and a pressure of 10 bar. Recently, experiments were performed to test a design modiﬁcation of the meter (Schieber, 2000). In Fig. 11.13, the mechanical accuracy Acc of the meter as obtained in experiments is plotted versus the Reynolds number. The low pressure curve is changing rapidly with increasing Reynolds number, approaching a constant value for higher Reynolds numbers. Starting from a certain critical Reynolds number, in this case about Re = 150, 000, the high pressure curve remains almost constant. This behavior agrees very well with the results of the twodimensional study described above.

11.7 Secondary Flow Due to the local ﬂow acceleration, classical secondary ﬂow patterns develop in the present conﬁguration. The induced secondary velocities in the cross-planes are relatively small, however, as the turbine is extremely lightly loaded. In Fig. 11.14, the contours of axial velocity are shown in a travers plane at the axial position x = −0.08 at the end of the stator guide vanes. A boundary layer is clearly visible at all solid surfaces. In the center of the channel, the axial velocity reaches its maximum. Due to the rapid change of the hub diameter, two secondary vortices develop in the lower corners. A much more detailed view of the ﬂow in the lower and upper corners reveals tertiary corner vortices at the hub and the outer casing.

11.8 Summary

199

Fig. 11.14. Secondary ﬂow in a travers plane at x=-0.08.

11.8 Summary The detailed study of the present numerical results indicated that ﬂow separation occured at the leading edge of the pressure side of the rotor for angles of incidence βi > 2.5◦ and on the suction side for all investigated angles of incidence 0◦ < βi < 5.0◦ . Vortices originated at the top and the bottom side of the trailing edge. The vortex formation depends on incoming Reynolds number. Vortex formation at the trailing edge affects the complete ﬂow ﬁeld around the rotor. The shift of the stagnation point and the change of circulation for different ﬂow capacities Q˙ conﬁrm this results. The high pressure shift of the rotor is shown to be attributed to a dependency of the ﬂow character on the Reynolds number. This idea was indicated by the results of two independent simulations and experiments. The conﬁguration of the three-dimensional model of the ﬁrst stage of the AccuTest TM turbine meter, consisting of approximately 1, 000, 000 cells, was presented. In order to simulate the unsteady three-dimensional ﬂow in a geometry including all details of the real meter, e.g., the tip gap, it was decided to simulate only the ﬁrst stage of the meter. The major results can be brieﬂy summed up as follows: a) The grid was sufﬁciently accurate to simulate the unsteady 3-D effects and to investigate the secondary ﬂow effects. b) Stator wakes cause large amplitude variation of coefﬁcient of moment c m of ﬁrst rotor in time. c) Boundary layers at casing and hub of the rotor increase the angle of incidence in the rotor. d) Non-axial ﬂow components due to the inlet geometry do not have strong effect on the ﬂow through the rotor. e) Flow effects that were found in two-dimensional simulations also occured in three dimensions. In future work there are some phenomena that should be further investigated to improve the accuracy of turbine ﬂow meters. It has been shown that the high pressure shift can be explained as a dependency of the ﬂow character on the Reynolds number. A critical Reynolds number Recrit , deﬁned as the Reynolds number at which a vortex separation from the trailing edge begins, exists. It is desirable to improve the design

200

11 Investigation of Unsteady Three-Dimensional Flow Fields in a Turbine Flow Meter

of the rotor to reduce this critical Reynolds number to values smaller than those occuring under any operating conditions. The inﬂuence of the wake of the ﬁrst stator on the ﬁrst rotor has been shown. In future investigations the axial gap between guide vanes and rotor and the design of the guide vanes could be reconsidered to reduce the extend and the effect of the wake on the incoming ﬂow ﬁeld in the ﬁrst rotor.

References Baker RC (1993) Turbine Flowmeters: II. Theoretical and Experimental published Information. Flow Meas Instrum, 4(3):123–144 Bonﬁg KW (1977) Technische Durchﬂußmessung. Vulkan-Verlag, Essen Brümmer A (1997) Der Einﬂuß von Druckpulsationen auf die Meßgenauigkeit von Turbinenradzählern. In: Workshop Kolbenverdichter, Rheine, 29.-30.10. Fluent (1998) Fluent 5, Lebanon, NH, USA Kallenberg M (1999) Numerische Simulation der chemisch reagierenden Strömung in einer typischen Überschallbrennkammer. Dissertation, Universität Essen Lee WFZ, Blakeslee DC, White RV (1982) A Self-Correcting and Self-Checking Gas Turbine Meter. Journal of Fluids Engineering, 104:143–149 Lee WFZ, Karlby H (1960) A Study of Viscosity Effect and Its Compensation on TurbineType Flowmeters. Journal of Basic Engineering, Transactions ASME, 82:717–728 Lehmann N (1990) Dynamisches Verhalten von Turbinenradgaszählern. Das Gas- und Wasserfach, 131(4):160–167 Mickan B et al. (1996) Die Fehlerverschiebung eines Turbinenradgaszählers in Abhängigkeit vom Anströmproﬁl. PTB-Mitteilungen, 106(2):113–124 Perpeet S (2000) Numerische Simulation von Strömungsfeldern um DurchﬂußMeßanordnungen. Dissertation, Universität Essen Schieber W (1998) A Turbine Meter with Built-in Transfer Prover. In Proceedings of the 9th International Conference on Flow Measurement, FLOMEKO ’98, pages 509–516, Lund, Sweden, June Schieber W (2000) United States Patent, No. 6,065,352. American Meter Company Yao J (1997) Numerische Simulation der Multi-Dimensionalen Strömung in Kolbenmotoren. Dissertation, Universität Essen

12 How to Design a New Flow Meter from Scratch Franz Peters, Carsten Ruppel Institut für Strömungslehre, Universität Essen, 45117 Essen, Germany

The idea of this chapter is to base a new pipe ﬂow meter on known ﬂuid mechanics and mechanics from the beginning rather than to explore the underlying physics after the event. Our new method relies on the well-known drag force on a cylindrical body and the force measurement by a load cell (Drag Force Flow Meter: DFFM). By means of an upstream screen the meter is prepared to axial velocity proﬁle distortions and, by the way the force is measured, swirl is invalidated. Tests with various pipe installations demonstrate the main features and the error bar to be expected.

12.1 Introduction Most of this book deals with the ﬂuid mechanics and signal processing of existing ﬂow metering techniques. In this chapter we want to turn the perspective around. We want to design a new meter from scratch. The meter is to be based on known ﬂuid mechanics and known force measurement and the installation problem is to be included in the design from the beginning. At the end of the design process the meter characteristics are to be clear and available with no need for retrospective research. Looking for a measuring principle (Ruppel and Peters, 2001) we observed that the force is hardly used among the numerous measured quantities like length (rotameter), time (ultrasound), frequency (vortex meter), pressure (oriﬁce, Venturi) or electrical potential (electromagnetic ﬂow meters). At the same time force measurement is not at all unusual in ﬂuid mechanics, especially aerodynamics, although forces can be small down to the Newton range. We choose a thin bar (a cylinder in this investigation) mounted radially from wall to wall in a pipe’s cross section and record its drag force which is a measure for the velocity. Normal ﬂow past a cylinder is one of the most fundamental ﬂow cases quoted as the drag coefﬁcient vs Reynolds number plot in most ﬂuid mechanics text books (Schlichting, 1960). We have to take into account though that the cylinder collects a velocity proﬁle along the radius rather than a uniform velocity. We also have to consider that in real situations the proﬁle may not be rotationally symmetric and that the ﬂow may be disturbed in the way of swirl. These problems are successfully

202

12 How to Design a New Flow Meter from Scratch

solved by an upstream screen that ﬂatens the proﬁle and by the way the cylinder is mounted which invalidates the swirl. Due to expense of time and material the experimental part of the following could only be a case study for gas in a 150 mm i.d. pipe featuring an 8 mm cylinder. Yet, similarity is based on Reynolds number and Dean number which lie in a realistic range. In principle the extension to liquids should be feasible and worthwhile.

12.2 Measuring Principle We divide the description of the measuring principle into two parts. The ﬁrst part deals with the problem how the ﬂow exerts a force onto the cylinder and how this force relates to the ﬂow rate. The second part shows how this force is properly measured. 12.2.1 Flow Rate and Drag Force A body in a uniform stream of velocity u and ﬂuid density ρ experiences a drag force which is proportional to r, the projection area A in ﬂow direction and the square of the velocity. The proportionality coefﬁcient, called drag coefﬁcient c D , depends on the Reynolds number Re (12.1) cD = cD (Re) u dcyl (12.2) ν based on the dynamic viscosity ν and a characteristic length of the body which is the cylinder diameter dcyl in our case. The most frequently quoted data on cD (Re) of a cylinder are those by Wieselsberger (1921, 1922) reproduced in Fig. 12.1. When the cylinder is mounted radially across the pipe as shown in Fig. 12.2 ﬂow direction and magnitude at the cylinder will in general depend on the radius r and the Re =

10

cD 1

0.1 0 10

101

102

103

104

Re

106

Fig. 12.1. Drag coefﬁcient of a cylinder as function of Reynolds number from Wieselsberger (1921, 1922).

12.2 Measuring Principle load cell

plate

force F cover

density U

D = 150 mm

grid

cylinder

r idealized force-profile

dZyl = 8 mm

F(r)

203

2D

flange

idealized velocity-profile velocity-profile u(r,M)

u(r,M)=um

Fig. 12.2. Working principle of the Drag Force Flow Meter (DFFM)

angle ϕ. In well-deﬁned cases such as developed laminar ﬂow, developed turbulent ﬂow and uniform ﬂow (square proﬁle) the ﬂow remains normal to the cylinder in pipe direction and the velocity is merely a function of the radius. In general, reality differs considerably from these ﬂow patterns, yet they lend themselves for reference purposes. In reality the proﬁles show distorted shapes and swirl occurs superposed to the main ﬂow not to speak of large scale turbulence which may add time dependence. A key beneﬁt of our design is that we do not need to worry about the swirl component from the beginning. The way the beam is mounted (see below) only the force in pipe direction is measured meaning that only ﬂow in pipe direction is recorded which is exactly the one that contributes to the ﬂow rate. Velocity and Reynolds number along the cylinder at a ﬁxed angle ϕ are now functions of the radius u(r) dcyl (12.3) Re(r) = ν so that the drag coefﬁcient has to be written in the form cD = cD (Re(r))

(12.4)

yielding the total force F on the cylinder in ﬂow direction (D = 2R) 1 F = ρ dcyl 2

R cD (Re(r)) u2 (r) dr .

(12.5)

−R

In the simple reference case of the square proﬁle with uniform velocity u m the force reduces to (12.6) Fsq = ρ dcyl cD (Rem ) u2m R . Using this reference force Fsq for scaling the general force of (12.5) we have

204

12 How to Design a New Flow Meter from Scratch

1 F = Fsq 2

1 −1

cD (Re(r)) u2 (r) r d . cD (Rem ) u2m R

(12.7)

We now interpret the integral as a shape factor f 1 f= 2

1 −1

cD (Re(r)) u2 (r) r d . cD (Rem ) u2m R

(12.8)

The shape factor is a measure for the deviation of the real proﬁle with respect to the reference square proﬁle including the inﬂuence of a variable drag coefﬁcient along the cylinder. Figure 12.1 teaches that the drag coefﬁcient is approximately constant over a wide range of Reynolds numbers which is limited by very small (Re < 500) and very large Reynolds numbers (Re > 2 × 105 ). Reynolds numbers > 2 × 105 are not envisaged in our design. Reynolds numbers down to zero are present in the wall boundary layer, yet the contribution of this range is very small anyhow. Therefore we approximate (12.9) cD (Re(r)) ≈ cD (Rem ) . With this approximation the shape factor can readily be calculated for known proﬁles. For the parabolic laminar proﬁle we ﬁnd f = 2.13 independent of Reynolds number, which means that the laminar proﬁle more than doubles the force as compared to the square one. This shows the dominant inﬂuence of the squared velocity in a peaked proﬁle. The shape factor of the developed turbulent proﬁle exhibits a slight dependence on pipe Reynolds number ReD as Table 12.1 shows. For the calculation the proﬁle given by Gersten and Herwig (1992) was used. In order to express the mass ﬂux m ˙ as function of the force F we eliminate u m from (12.6) and in the resulting equation replace Fsq by means of (F/f ) from (12.7, 12.8). Then multiplication with the density ρ and the pipe area R 2 π yields R3 π 2 ' 1 Fρ. (12.10) m ˙ = cD (Rem ) f dcyl Table 12.1. Shape factor f for developed turbulent proﬁles for various pipe ﬂow Reynolds numbers ReD ReD =

um D ν

50,000 100,000 150,000 200,000 250,000

shape factor f turbulent proﬁle 1.170 1.158 1.151 1.146 1.142

12.2 Measuring Principle

205

Here the force is the measured quantity. The density is either known or measured through temperature and pressure for a gas. Pipe and cylinder diameter are given. The drag coefﬁcient is close to 1 (Fig. 12.1). The crucial parameter is the form factor f . It is only known when the proﬁle is ideal (Table 12.1) or measured which, of course, would save the ﬂow meter. Our answer to the real, distorted proﬁle is a wire screen placed upstream of the cylinder. Such a screen produces relatively small pressure losses and needs little space. It has a smoothing and equilibrating effect on upstream proﬁle distortions such that the downstream proﬁle approaches the square shape. In contrast to the ﬂow straightener it has little impact on the swirl which we are not concerned about anyhow. Squaring the proﬁle the screen pushes the shape factor f closer to one without reaching it, of course. It gets close enough though, so that (12.10) is suitable for the design process of the meter which will normally start with the range of m ˙ required by the user for a given pipe of radius R. The cylinder ought to be much smaller than the pipe, dcyl D. With the product of cD and f close to one the force F on the cylinder follows for kown density. F then determines the design of the load cell. After the design has been completed and the load cell has been made calibration of m ˙ vs F ρ according to (12.10) in a reference situation becomes inevitable to accomplish high accuracy. We’ll go through these steps further down concluding with an investigation on the sensitivity to installations. 12.2.2 Drag Force and Load Cell

100 55

load cell

R3

w

4

15

t=15 15

cylinder

Fig. 12.3. Design of the load cell with (right) and without load (left)

H

The ﬂuid force on the cylinder is an integrated load with varying load distribution. To measure this load independent of the distribution a load cell as shown in Fig. 12.3 is suitable. Kinematically the four thin portions of the cell constitute the joints of a parallelogram. The front part with the attached cylinder moves under load parallel to the rear part ﬁxed to a rigid base such that the attached cylinder remains in radial direction in the pipe. At the same time the thin portions ﬂex like leaf springs transforming the acting force into deformation. The deformation can be measured by strain gauges or optical displacement sensors (or others). Our goal is the design of the entire meter. Therefore, as an example, we selected material (aluminum 7075 T7351) and shape of a particular load cell as given in Fig.12.3. All measures but the wall thickness w of the ﬂexible joints were ﬁxed. This given conﬁguration was fed into a FEM program to calculate the displacement

F

206

12 How to Design a New Flow Meter from Scratch

of the cylinder as function of the load for various wall thicknesses. The calculated results were checked versus a simple measurement in which the load was simulated by weights and the displacement was measured by an optical gauge (optoNCDT by Micro-Epsilon). Excellent agreement within ±2.5% was found. The outcome of this investigation is the following formula ε w2.7935 = 0.0982 F .

(12.11)

It relates the wall thickness w in mm, the displacement ε in mm and the force F in Newton. Most importantly, a linear relationship between force and displacement is achieved. In the design process the maximum force will be known from the ﬂow rate as described above. A corresponding maximum displacement has to be chosen according to the type of length measurement available (strain gauges, optical devices or others). Then (12.11) delivers the thickness w, which of course is bound by upper and lower limits. The lower limit is set by machining and the upper limit by the condition that the ﬂexible joint needs to be thin relative to the side-pieces of the cell. The range 0, 5 mm < w < 1, 5 mm was found reasonable for the test design. Please note that this test design was just to show the designability of the load cell. The ranges of measures and forces have to be tuned to the application.

12.3 Setup and Calibration Figure 12.4 illustrates the basic setup in the calibration mode. An air blower driven by a frequency controlled AC motor circulates room air through a pipe system (precision pipes DIN 2391). Air enters the test pipe of D = 150 mm i.d. through a nozzle and a wire screen of 1 mm wire and 3 mm mesh size. The pipe is segmented and assembled by ﬂanges (not shown) to allow variable conﬁgurations of pipe and installations including the DFFM and the screen which are integrated into ﬂanges themselves. The big pipe reduces to a smaller one of d = 100 mm i.d. which incorporates a Venturi meter 20 d upstream of the blower. The Venturi provides the mass ﬂux reference after careful calibration against another reference placed downstream of the blower described elsewhere (Wildemann, 2000). The DFFM is ﬁtted with a 8 mm cylinder of carbon ﬁbre laminate for best stiffness to mass ratio. The displacement was to be measured by an integrated strain gauge because we were not able to 20 d

34 d

14 D

2D

blower

m

reducer

D

d

m venturi meter

DFFM

inlet nozzle

installation

Fig. 12.4. Calibration setup indicating installation range

12.3 Setup and Calibration

207

integrate the optical device mentioned earlier into the ﬂange. A commercial load cell (PW4FC3 by HBM) of the type tested in Sect. 12.2 was purchased in one piece with the strain gauge and an appropriate ampliﬁer (BA 651 by BLH). Data aquisition is by a National Instruments A/D card “PCI-6024E” in combination with the LabView 6i software. Besides the force temperature and pressure in the pipe are recorded at the sampling rate of 104 s−1 . Due to an eigen vibration of the bar/cell arrangement of the order of 200 Hz a mean value of the force has to be taken. In order to do so the integration time should not drop below 0.1 s. The LabView program views the recorded data including the reference mass ﬂux from the Venturi. The calibration procedure is as follows. Starting from u m = 3.5 m/s mean velocity in the big pipe the blower motor frequency is risen stepwise up to the mean pipe velocity of about 32 m/s. For each step ρ is determined from p and T (including relative humidity) and F is measured. The reference mass ﬂux m ˙ is obtained from the Venturi. This information together with (12.10) yields the calibration diagram Fig.12.5 displaying cD f versus Re. As cD is close to one (see Fig.12.1 and discussion) the message drawn from this plot is that f in fact comes close to one too meaning that the screen provides an approximately square proﬁle at least in this calibration arrangement right downstream of a rounded inlet nozzle. Of course, the residual Reynolds number dependence must be taken into account. The relationship cD f = 0.64 exp(−64 × 10−5 Re) cos(−535 × 10−6 Re − 0.05) +1.91 × 10−6 Re + 0.997

(12.12)

ﬁts the data quite well constituting the analytical calibration curve of this particular DFFM. With the help of (12.10) it transforms into the mass ﬂux vs force relationship.

1.1

cD·f 1 0.95 0.9 0.85 0.8

0

2500

5000

7500

10000

12500

Fig. 12.5. Calibration curve

Re

17500

208

12 How to Design a New Flow Meter from Scratch

12.4 Results for Installations A calibrated DFFM exposed to any square velocity proﬁle in the adequate range will be accurate within a small margin of reproducibility. Obviously, it is not possible to predict the efﬁciency of the screen for any installed upstream disturbance. As a consequence typical installation situations have to be tested before a reliable error prediction can be made.

Fig. 12.6. The investigated installations

Figure 12.6 compiles seven variants covering quite common installations and an extreme one. The common ones are the inlet nozzle with 10 D of straight pipe (A), two regular elbow bends (B,C) and two double bends (D,E). A pipe expansion (F,G) opening from 100 to 150 mm i.d. on a short distance of 3.3 D introduces ﬂow separation, an undoubtedly extreme disturbance bound to lead to substantial indication errors of most any meter placed close to it. Results are compiled in Fig. 12.7. The relative deviation of the DFFM in terms of ˙ ref m ˙ DFFM − m 100 (12.13) ∆m ˙ = m ˙ ref is plotted versus the reference mass ﬂux. The covered range of mass ﬂuxes corresponds to mean velocities in the big pipe from 5 to 28 m/s. When ∆m ˙ > 0 the DFFM overpredicts the ﬂow rate and the other way round for ∆m ˙ < 0. The ﬁrst ﬁve installations show consistent negative deviations with some scatter. In view of the shape factor discussion the consistency suggests that in all these cases the proﬁle tends to be somewhat “ﬂater” than in the calibration case. Elbow and double elbow produce strong disturbances which are described by Hilgenstock and Ernst (1996). Generally, the main problem of the disturbance is swirl which is known to decay slowly (Schlüter and Merzkirch, 1996). Here the DFFM is in the advantageous position that it is not affected by radial and tangential velocity

12.5 Conclusions

209

5

A B C D E F G

'm (%) 3 2 1 0 -1 -2

0

100

200

300

400

mRef (g/s)

600

Fig. 12.7. Relative deviation of the mass ﬂux measured by the DFFM versus the reference mass ﬂux for the installations of Fig.12.6.

components as already explained in Sect. 12.2.2. What remains is a moderate dent in the proﬁle which is dealt with by the screen and, to some extent, by the 10 D of pipe length. In sum, the installations lead to deviations from the calibration arrangement, yet, it must be emphasized that the deviation stays within a percent! Furthermore the deviation exhibits a systematic character which could be corrected for. The pipe expansion of installation F,G produces substantial positive deviations. Certainly, it is not recommendable to place a DFFM (or any other instrument) into such a position. We have included these installations just to show how well the deviations of a DFFM can be interpreted. The diffusor angle of the expansion enforces ﬂow separation with high velocities on the center of the pipe and low ones in the separated region (or even negative). The proﬁle will be extremely peaked entailing large positive deviation up to 4% in case G. In the other case F with the DFFM moved downstream the ﬂow seems to be reattached with a broader proﬁle and correspondingly smaller deviations down to half a percent for the higher ﬂow rates.

12.5 Conclusions We have shown how to design a ﬂow meter for the mass ﬂux in pipes from scratch. The design is based on fundamental ﬂuid mechanics (ﬂow past a cylinder) and a well understood type of load cell. The load cell determines the force on the cylinder which is an accurate measure of the ﬂow rate when the velocity proﬁle is known. As the proﬁle is normally not known the DFFM generates its own proﬁle by means of an upstream screen. This proﬁle approximates the square one used for calibration. The deviation of the screen proﬁle from the calibration proﬁle causes the uncertainty of the measurement. Tangential and radial disturbances in the velocity distribution are ignored by the meter from the beginning saving further consideration. In a number of real installation situations different from the calibration situation we have demonstrated that the DFFM works very well. It shows an error bar of less

210

12 How to Design a New Flow Meter from Scratch

than a percent with a consistent shift that can be rationalized so that accuracy could be even improved by corrections. The elements of the meter allow reﬁnements which should be looked at in further developments. Especially, the selection of the screen parameters has some potential. Worth mentioning is that two screens, upstream and downstream, would allow to reverse the ﬂow direction. As the measuring signal, the drag, is based on Reynolds number and as the dimensions of the load cell can be calculated the meter can be laid out for most ﬂow metering situations.

References Gersten K, Herwig H (1992) Strömungsmechanik - Grundlagen der Impuls-, Wärme- und Stoffübertragung aus asymptotischer Sicht. Vieweg-Verlag, Braunschweig, Wiesbaden Hilgenstock A, Ernst R (1996) Analysis of installation effects by means of computational ﬂuid dynamics-CFD vs experiments?. Flow Meas.Instrum. 7:161–171 Ruppel C, Peters F (2001) Messung des Massenstroms in Rohrleitungen durch die Widerstandskraft eines quer angeströmten Zylinders (DFFM-Methode). tm-Technisches Messen 68:465–472 Schlichting H (1960) Boundary Layer Theory. 4th ed. McGraw-Hill, New York Schlüter T, Merzkirch W (1996) PIV measurements of the time-averaged ﬂow velocity downstream of ﬂow conditioners in a pipeline. Flow Meas Instrum 7:173–179 Wieselsberger C (1921) Neuere Feststellungen über die Gesetze des Flüssigkeits- und Luftwiderstandes. Physikalische Zeitschrift 22:321–328 Wieselsberger C (1922) Weitere Feststellungen über die Gesetze des Flüssigkeits- und Luftwiderstandes. Physikalische Zeitschrift 23:219–224 Wildemann C (2000) Ein System zur automatischen Korrektur der Messabweichung von Durchﬂussmessgeräten bei gestörter Anströmung. Doctor thesis, Universität Essen. See also: Fortschritt-Berichte VDI, Reihe 8, Nr. 868

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body Ernst von Lavante Institut für Strömungsmaschinen, Universität Essen, 45117 Essen, Germany

In the present chapter, the problem of accurate determination of mass ﬂows by means of the so-called vortex-shedding ﬂow meter is studied for the case of disturbed inﬂow. To this end, the ﬂow around bluff bodies of different designs, used in typical vortexshedding ﬂow meters, was numerically simulated using a solver of the unsteady, compressible Navier–Stokes equations in three dimensions. The computations were carried out for several types of disturbed inﬂow conditions, including bends, ﬂow with rotation oriented in the axial direction and unsteady ﬂow ﬂuctuations. The results were compared with experimental data obtained in an ongoing investigation. The effects of turbulence were modeled by using large-eddy simulation (LES). The resulting ﬂow ﬁelds were analyzed using various methods, including visualization, evaluation of several of the features and DFT of properly chosen variables. Recommendations regarding the operational conditions of the vortex-shedding ﬂow meters are made.

13.1 Introduction Many aerodynamic problems require a volume- or mass-ﬂow data for its quantitative solution. Therefore, a number of methods for ﬂow rate measurement have been developed. One relatively simple and promissing ﬂow measurement device is the so-called vortex-shedding ﬂow meter, in which the mass ﬂow is determined by observing the relationship between vortex-shedding frequency from a bluff body attached inside a channel, and the corresponding mean velocity about it. The bluff body causes the production of a system of periodic vortices (“vortex street”), whose frequency can be correlated with the mean ﬂow velocity and, therefore, the mass ﬂow. This procedure assumes a regular and well-deﬁned vortex structure as well as shedding mechanism, resulting mostly in linear dependency of the mass ﬂow on the shedding frequency over a wide range of Reynolds numbers. However, it has been observed that certain types of disturbancies in the inﬂow result in rather irregular pressure signatures of the vortex system or even shift of its charachteristic

212

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body

Fig. 13.1. Principle of a vortex-shedding ﬂow meter.

frequencies, leading to unreliable mass ﬂow data. A schematic picture of a typical vortex-shedding ﬂow meter is offered in Fig. 13.1. Commercial ﬂow meters use a large variety of bluff body shapes, often restricted by the attachment of the pressure sensors or, more likely, patent laws. Previous application of some current bluff body designs lead to fairly irregular pressure signatures, making them unreliable. It was, therefore, decided to investigate not only the known bluff body shapes, but also a few alternate designs, developed empirically by the present authors. In the past, various shapes of vortex bodies have been tested with regard to their applicability to a simple signal processing. Triangular shapes (Johnson, 1990; Fureby, 1995; Madabhushi et al., 1997; Hans et al., 1998) as well as shapes with truncated tips (e.g., Hans et al., 1997) have been tested. In the preliminary part of this work, tests were performed with T-shaped, rectangular and a new-designed bluff body. Well-deﬁned vortices were generated, giving, after signal processing, excellent measurement results (Hans et al., 1998; Miau et al., 1993). A strong and well-deﬁned dependency of the vortex frequency on the ﬂow velocity and, therefore the Reynolds number, could be obtained. The theory of operation is treated in more detail in Chaps. 5 and 6. In the work concerning the vortex shedding ﬂow meters included in this book, the method of detection of the vortices that has been employed was always the same. Downstream of the bluff body, an ultrasound barrier was installed, “counting” the vortices. In the numerical simulation, this signal was approximated by an integral of the vertical velocity component over an area closely resembling the ultrasound beam. In the present work, the ﬂow was investigated using a Navier–Stokes equations solver, capable of handling unsteady, compressible and viscous ﬂows over a wide range of Mach numbers. Several different classes of inﬂow disturbancies were studied and analyzed in regards to their effects on the ﬂow metering capabilities.

13.2 Numerical Algorithm The numerical algorithm employed uses the three-dimensional, time-dependent full Navier–Stokes equations describing the conservation of mass, momentum and energy of ﬂuid ﬂows. The divergence form in body-ﬁtted, curvilinear coordinates is:

13.2 Numerical Algorithm

∂Q ∂(F − Fv ) ∂(G − Gv ) ∂(H − Hv ) + + + =0, ∂τ ∂ξ ∂η ∂ζ

213

(13.1)

with Q = J −1 (ρ ρu ρv ρw e)T the vector of the conserved variables. J is the Jacobian of the coordinates transformation from physical (x, y, z, t) to computational (ξ, η, ζ, τ) space. The program is based on the ﬁnite-volume formulation, using a cell-centered organization of the control-volumes. The spatial discretization is carried out with the help of Roe’s Flux Difference Splitting Scheme, a Godunov-typ method providing an approximate solution of the Riemann problem on the cell interfaces. Here, the ﬂux is (Vatsa et al., 1987): 0 1/ F (Q)L + F (Q)R + | A˜i+ 21 | (QR − QL ) . (13.2) Fi+ 12 = 2 The index i + 12 describes the values on the cell interfaces, R and L the values right and left from it. The Roe-averaged matrix A˜ is given by the differentiation of the local-linearized function F (Q). This scheme is formally central difference plus a damping term. The method has been proved to be very accurate and effective in the simulation of low Mach number viscous ﬂows (von Lavante and Yao, 1993). Upwind-biased differences are used for the convective terms, central differences for the viscous ﬂuxes. Starting with a constant initialization of the scalar variables and body-ﬁtted velocity components, the integration in time is carried out by a modiﬁed explicit Runge–Kutta time stepping as well as, optionally, an implicit Approximate-Factorization method or symmetric Gauss–Seidel scheme (Jameson et al., 1981). 13.2.1 Low Mach Number Modiﬁcations Numerical algorithms for the simulation of compressible ﬂows become inefﬁcient and inaccurate at very low Mach numbers. The difﬁculties are due to the formulation of the governing equations in their discretized form. The problems have been addressed by many previous investigators, including Shuen et al. (1993), Pletcher and Chen (1993) and Edwards and Roy (1998). The main problem is the stiffness of the governing equations at low Mach numbers. The condition number of the Jacobian matrices (ratio of the maximal to the minimal eigenvalues) increases to inﬁnity as the Mach number approaches zero. The time marching step therefore is restricted because of the large disparity of the eigenvalues of the Jacobian, representing the convective and acoustic signal speeds, respectively. The second problem results from the pressure singularity in the momentum equations at very low Mach numbers. The ratio of the magnitudes among the pressure term p and the convective terms ρu 2 is inversely proportional to the Mach number squared M 2 . The large difference in magnitude will yield a large roundoff error. To enable efﬁcient numerical solutions of the equation system at low Mach number, a pseudo-time term was added to the time-dependent compressible Navier– Stokes equations. The primitive variables were employed as unknowns, rather than the traditional conservative variables. A preconditioning matrix Γ was used in order

214

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body

to eliminate the time-step difference between the convective and acoustic characteristic speeds at low Mach number. The preconditioned equations can be discretized in the delta form: 2 1 ∂(AˆH ) ∂(AˆF ) ∆τ ˆ ˆ= AQ + ∆τ + ... + ∆Q Γ+ ∆t ∂ξ ∂ζ − ∆τ

∂H ∂Q ∂F + + ...+ ∂t ∂ξ ∂ζ

,

(13.3)

where the Jacobian matrices are deﬁned as, ∂F ∂G ∂H ∂Q , AˆF = , AˆG = , AˆH = . AˆQ = ˆ ˆ ˆ ˆ ∂Q ∂Q ∂Q ∂Q

(13.4)

ˆ is the vector of the primitive Here Q is the vector of the conservative variables and Q T variables (p, u, v, w, T ) . A more detailed description of the matrix Γ can be found in Yao and von Lavante (1997). To obtain time-accurate solutions, a dual time-stepping integration procedure was used. For each physical time step, the equations were integrated in pseudo-time. The physical time term behaves like a source term during the pseudo-time iterations. The converged solution in pseudo-time corresponds to a time-accurate solution in physical time. For steady ﬂow problems, the physical time terms are neglected. An implicit iterative procedure using symmetric Gauss–Seidel scheme was used for the sub-iterations of the equations in the pseudo-time. The converged equations at each physical time step satisfy the original discretized Navier–Stokes equations. To circumvent the problem of large difference in magnitude of the convective and pressure terms in the momentum equations, a gradient splitting of the Euler ﬂux into convective terms and a pressure term can be considered: ∇(ρu2 + p) = ∇(ρu2 ) + ∇p.

(13.5)

Both split gradients are of the same order of magnitude, independent of the Mach number. The Liou’s Advection Upwind Splitting Method was applied to treat the convective and pressure terms separately. The convective terms were upstreambiased using an appropriately deﬁned advection Mach number at the cell interface, while the pressure term was strictly dealt with by using acoustic waves. One of the advantages of the method is that the upwind effect can be easily achieved with few modiﬁcations of the programs. 13.2.2 Turbulence Model The highly complex geometrical and physical conﬁgurations are making the choise of an appropriate turbulence model very difﬁcult. In the vortical, periodically separating ﬂow with strong local rates of acceleration, any isotropic models are not adequate. The present authors decided, therefore, the use of the concept of large eddy simulation (LES).

13.2 Numerical Algorithm

215

13.2.3 Veriﬁcation The present numerical algorithm was subjected to veriﬁcation of it’s temporal and spatial accuracy and consistancy. The scheme is formally second order accurate in space, since the viscous terms are obtained from second order central differences. The scheme was ﬁrst veriﬁed using the usual grid reﬁnement study for the case of viscous ﬂat plate ﬂow at a free stream Mach number of M∞ = 0.5. Deﬁning the global error as the L2 - norm of the deviation of the present solution from the Blasius solution, second order accuracy was veriﬁed (von Lavante, 1990). Next, the combination of the present solution scheme with the computational grid was investigated for the case of an optimized bluff body shape using three different grids. Figure 13.2 shows the density plot and summarizes the results. It should be noted that the ﬁnest grid very closely approaches the Strouhal number of the corresponding experiment.

Sr 0.28 0.275 0.27 0.265 0.26

- numerical

0.255

- experimental 0.25

0

12500

25000 37500 Meshpoints

50000

Fig. 13.2. Grid reﬁnement study for an optimized bluff body shape.

The temporal accuracy was tested by simulating the 2-D ﬂow about a cylinder, with free stream Mach number M∞ = 0.1 and a Reynolds number of Re = 200. The resulting Strouhal number of the vortex separation is Sr = 0.196 and therefore within the range given in literature.

216

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body

13.3 Results The bluff body inside the vortex-shedding ﬂow meter causes separation of periodic vortices from the upper and lower edge of a bluff body, similar to the well-known von Kármán vortex street. In the present effort to optimize the bluff body, many different shapes were testet. As a result, an “optimum” shape was found and extensively tested (Hans et al., 1998) in 2-D, 3-D and experimentally. The resulting plot of the frequency of vortex passage through an ultrasound barrier placed downstream of the bluff body is displayed in Fig. 13.3 as a function of the time averaged mean velocity in the pipe. Its highly linear behavior in the given velocity range is obvious. Somewhat surprising is the good agreement between the 2-D and 3-D numerical results. The right side of Fig. 13.3 shows the shape of the optimized bluff body.

Fig. 13.3. Frequency vs. pipe mean velocity and optimized bluff body shape.

Figure 13.4 shows two basic strategies of mounting the bluff body inside a vortex-shedding ﬂow meter. The left device causes the separation of periodic vortices from the upper and lower edge of a bluff body, similar to the well-known von Kármán vortex street. The right one produces axially-symmetric rings in the wake of a bluff body.

Fig. 13.4. Two different designs of bluff body mounting inside a pipe.

13.3 Results

217

Fig. 13.5. Pressure Contours in a single bend with the vortex ﬂow meter at the lower end.

As long as the inﬂow remains undisturbed, the shedding frequency in the wake of the bluff body is rather constant for most shapes. There are many studies concerning the optimization of the bluff body shapes, investigated under the condition of undisturbed inﬂow. We decided to examine the development of the vortex-shedding frequency in a more practical view. Flow meters are normally part of a pipe system and therefore the inﬂow is often nonuniform or even unsteady. Figure 13.5 shows a section of the pipe, with a single bend, featuring the protruding bluff body downstream of the bend, visible at the lower end of the picture. Figure 13.6 displays the vortex ﬂow meter with the double bend. In the case, the meter is at the upper end of

Fig. 13.6. Pressure Contours in a double bend with the vortex ﬂow meter at the upper end.

218

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body

the pipe. The characteristic double vortex system in the wake of the bend can to be seen. Figure 13.7 left shows the result of DFT analysis of the ﬂow behind the bluff body. The result is shown after introducing a single bend upstream of the meter, displaying a very strong peak at the meter’s natural vortex-shedding frequency. In this case, the dominating frequency remains unchanged. Figure 13.7 right shows the DFT analysis of ﬂow past a double bend. Again, a very strong peak at the meter’s frequency is noticeable.

sk

sk

1

1

0.9

0.9

0.8

0.8

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

f [Hz]

0 -0.1

0.1

f [Hz]

0 0

500

1000

1500

-0.1

0

500

1000

1500

Fig. 13.7. Shedding frequency for ﬂow downstream a single bend (left) and with a double bend upstream the meter (right).

The signal in Fig. 13.7 was obtained by observing the modulation of the radial velocity component, as given by the present conﬁguration with the ultrasound barrier, described in more detail in Chap. 6. However, the pressure ﬁeld in the single 90 ◦ -bend was highly unsteady, with a wave that was alternativly reﬂected between the lower and upper wall while rotating around the pipe axis and slowly moving upstream (Fig. 13.8, to the top). Table 13.1 summarizes the results. The Strouhal number remains almost constant for all introduced bends. Due to the large tangential velocities in magnitude of the pipe mean velocity the deviation in case of the swirl is noticeable higher. Table 13.1. Results Swirl Meshpoints Reynolds number Frequency (Hz) v (m/s) Strouhal number

90◦ -Bend Double Standard horizontal vertical Bend Tube

Exp.

414300 621400 621400 621400 1.07 Mio. 183000 180000 173000 213000 191000 175000 287 300 295 348 307 280 27.7 27.29 26.18 32.18 28.9 26.48 0.249 0.264 0.27 0.26 0.255 0.254

13.3 Results

219

Fig. 13.8. Density waves in the single 90◦ -bend.

Finally, the effects of pulsating inﬂow on the accuracy and basic function of the vortex-shedding ﬂow meter were investigated. For this purpose, the two-dimensional computational model was modiﬁed by blocking the inﬂow portion of the pipe far upstream of the bluff body. The complete blockage was executed at a certain frequency for a small percentage of the period T corresponding to the frequency of pulsation. The ﬂow conditions were selected such that the undisturbed frequency of the meter was 250 Hz. The pulses were implemented at 50, 100 and 200 Hz, blocking the ﬂow for between 0.5% of T and 10% of T . At the lower frequency of pulsation (50 and 100 Hz), the vortex-shedding frequency was basically uneffected. In Fig. 13.9, the static pressure at the bluff body is plotted as a function of the number of iterations in time on the top of the picturefor the case of 100 Hz pulsation with the blockage lasting 0.5% of T . The pulses can be seen as sharp peaks in the pressure history, followed by the same periodic ﬂuctuation as for the undisturbed inﬂow. The simulated ultrasound signal, shown in the middle of the ﬁgure, is practically undisturbed. The mass ﬂow just after the blockage location is displayed at the bottom of Fig. 13.9. At the highest pulsation frequency of 200 Hz and 10.0% T duration, the situation is completely different. The disturbance frequency is much closer to the natural fre-

220

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body p

v

0.75

0.35 0.3 0.25

0.7

0.2 0.15 0.1

0.65

0.05 0 0.6

-0.05 -0.1 -0.15

0.55

-0.2 -0.25 0.5

0

25000

50000

75000

100000

125000

Iterations

Fig. 13.9. Pressure evolution and simulated signal for pulsation frequency of 100 Hz at 0.5% duration.

quency of 250 Hz, and the amplitude of the pressure disturbance is much higher due to the longer blockage time. As can be seen in Fig. 13.10, the pressure does not recover to its undisturbed ﬂuctuation, and the frequency of vortex shedding “locks” on the frequency of the pulsation. Needless to say, this behavior results in a catastrophic failure of the meter. v

p

0.45

0.75

0.4 0.35

0.7

0.3 0.25

0.65

0.2 0.15

0.6

0.1 0.05 0

0.55

-0.05 -0.1

0.5

-0.15 -0.2

0.45

-0.25 -0.3

0.4

0

15000

30000

45000

60000

-0.35

Iterations

Fig. 13.10. Pressure evolution and simulated signal for pulsation frequency of 200 Hz at 10.0% duration.

References

221

The results of the numerical simulation of pulsating inﬂow are summarized in Fig. 13.11, in which the frequency of the vortex shedding is shown as a function of the iterations in time and, therefore, also as a function of time.

Fig. 13.11. Frequency of vortex shedding with pulsating inﬂow.

After 50000 iterations, the pulsation was initiated. In the case 100 Hz pulsation signal, shown in Fig. 13.9, the frequency becomes somewhat irregular, but remains basically at 250 Hz. When the pulsation frequency is increased, however, to 200 Hz, and the time of the ﬂow blockage changed to 10% of T , the vortex-shedding frequency jumps to 200 Hz in the apparent “lock-in”.

13.4 Conclusions Several numerical investigations of unsteady and disturbed pipe ﬂows are presented. A veriﬁed and validated Navier–Stokes solver for three-dimensional simulations of compressible, viscous and unsteady ﬂow was applied producing results that were in good agreement with the corresponding experimental work. The simulations were useful in determining practical demands concerning the installation of the meter. The DFT analysis showed that the vortex-shedding ﬂow meter will still work properly under disturbed ﬂow conditions.

References Edwards JR, Roy CJ (1998) Preconditioned multigrid methods for two-dimensional combustion calculations at all speeds. AIAA Journal, 35, No.2:185–192 Fureby C (1995) Large-eddy simulation of turbulent anisochoric ﬂows. AIAA Journal, 33, No.7:1263–1272

222

13 Effects of Disturbed Inﬂow on Vortex Shedding from a Bluff Body

Hans V, Poppen G, Lavante E. v., Perpeet S (1997) Interaction between vortices and ultrasonic waves in vortex-shedding ﬂowmeters. FLUCOME ’97, Hayama Hans V, Poppen G, Lavante E. v., Perpeet S (1998) Vortex-shedding ﬂowmeters and ultrasound detection: signal processing and bluff body geometry. Flow Meas Instrum, 9:79–82 Jameson A, Schmidt W, Turkel E (1981) Numerical solutions of the Euler equations by ﬁnite volume methods using Runge–Kutta time-stepping schemes. AIAA Paper, 81-1259 Johnson MW (1990) Computation of ﬂow in a vortex-shedding ﬂowmeter. Flow Meas Instrum, 1:201–208 Madabhushi RK, Choi D, Barber TJ (1997) Unsteady simulations of turbulent ﬂow behind a triangular bluff body. AIAA Paper, 97-3182 Miau JJ, Yang CC, Chou JH, Lee KR (1993) A T-shaped vortex shedder for a vortex ﬂowmeter. Flow Meas Instrum, 4:259–267 Pletcher RH, Chen KH (1993) On solving the compressible Navier–Stokes equations for unsteady ﬂows at very low mach numbers. AIAA-paper 93-3368 Shuen JS, Chen KH, Choi Y (1993) A coupled implicit method for chemical non-equilibrium ﬂows at all speeds. J. of Comp. Phy. 106:306–318 Vatsa VN, Thomas JL, Wedan BW (1987) Navier–Stokes computations of prolate spheroids at angle of attack. AIAA Paper, 87-2627 von Lavante E (1990) The Accuracy of Upwind Schemes Applied to the Navier–Stokes Equations. AIAA Journal, 28, No. 7:1211–1212 von Lavante E, Yao J, (1993) Simulation of ﬂow in exhaust manifold of an reciprocating engine. AIAA 24th Fluid Dynamics Conference Yao J, von Lavante E (1997) Proceedings of 7th International Symposium on CFD:689–694, Beijing, PR China Hans V, Windorfer H, von Lavante E, Perpeet S (1998) Experimental and numerical optimization of acoustic signals associated with ultrasound measurement of vortex frequencies. FLOMEKO 98, Proceedings:363–367

14 Correction of the Reading of a Flow Meter in Pipe Flow Disturbed by Installation Effects Carsten Wildemann1 , Wolfgang Merzkirch1 , Klaus Gersten2 1 2

Institut für Strömungslehre, Universität Essen, 45117 Essen, Germany Institut für Thermo- und Fluiddynamik, Ruhr-Universität Bochum, 44780 Bochum, Germany

A method is described that allows the correction of the reading of a ﬂow meter exposed to pipe ﬂow disturbed by installations upstream of the meter. The method is aimed at minimising the distance between installation and meter and is based on characterizing the ﬂow by “fundamental” disturbances, physically interpretable as vortical structures, and on the detection of these disturbances by a measuring device located slightly upstream of the meter. A functional relationship between fundamental disturbances and “error shift” of the meter is postulated, and its existence is demonstrated by a number of experiments. It is concluded that the correction method is applicable to any type of installation and ﬂow meter.

14.1 Introduction Flow meters are usually calibrated in fully developed pipe ﬂow. For the practical use of a meter, existence of the same state of ﬂow at the position of the meter must be secured, and this requires providing certain lengths of straight pipe upstream and downstream of the meter, as speciﬁed in the technical norms. These requirements are often not met in practice; the meter is then exposed to ﬂow that is not fully developed, and measurement errors arise. Deviations from the fully developed ﬂow state in the pipe are caused by installations, e.g., bends, valves, junctions, etc. Two principal approaches are known for using the ﬂow meter with a minimum length of straight pipe between installation and meter, below the lengths prescribed by the norms. One is the use of a ﬂow conditioner that is supposed to accelerate the redevelopment of the ﬂow. The various types of disturbances of the fully developed velocity proﬁle caused by installations disappear at different decay rates downstream of a conditioner; see, e.g., Laws (1990), Xiong et al. (2003), and also Chap. 4 in this book. The second approach is to calibrate the meter in the presence of the speciﬁc installation. This possibility was investigated systematically for oriﬁce meters (Mattingly and Yeh (1994); Reader-Harris et al. (1995)). The velocity distribution downstream of various installations was measured and characterized. The reading of the

224

14 Correction of the Reading of a Flow Meter

oriﬁce was then corrected as a function of chararcteristic numbers describing the deviation of the disturbed velocity proﬁle from the fully developed state. While these approaches are restricted to the speciﬁc types of installation and meter used in these procedures, we describe here a method of correction that is aimed at being more universal, independent of the knowledge about the speciﬁc installations and, in principle, applicable to any kind of ﬂow meter. Our approach is based on the theoretical results of Gersten and Papenfuss as presented in Chaps. 2 and 3 of this book. It is shown there that any disturbance caused by an installation is composed of a number of “fundamental” disturbances that decay downstream of the installation according to different decay laws. The local (in axial direction) values of the magnitude of the fundamental disturbances characterize the local state of ﬂow, i.e., the deviation from the fully developed state. This is equivalent to describing the state of ﬂow by characteristic numbers, each of them representing one of the fundamental disturbances, see Chaps. 2 and 3 of this book. In contrast to the characteristic numbers as they had been used, e.g., by Mattingly and Yeh (1991) or Mickan et al. (1996), the fundamental disturbances represent speciﬁc structures in the ﬂow, i.e., they have a real physical signiﬁcance; see Chap. 2. The idea is now that the state of the ﬂow, expressed by a set of characteristic numbers, is measured slightly upstream of the ﬂow meter, and that the reading of the meter can be corrected if a relationship between this reading and the set of characteristic numbers is known. Knowledge of this relationship is provided by a calibration: The difference between the discharge coefﬁcient of the meter in fully developed ﬂow, CD∞ (deﬁnition see below), and the discharge coefﬁcient in disturbed ﬂow, CD , at the same value of the volumetric ﬂow rate, is measured for a limited number of disturbances and ﬂow rates. It must be proven that, once the relationship is established from this set of ﬁnite data values, the correction can be performed for any kind of disturbances caused by installations. The relationship between CD , characteristic numbers, and volumetric ﬂow rate (or Reynolds number) must be determined for each type of meter to be used. We apply the procedure to the Venturi and oriﬁce ﬂow meter. The deviation of the ﬂow proﬁle from the fully developed state is determined by measuring the azimuthal distribution of the wall shear stress along a circumference of the inner pipe wall, a short distance (< 1 D) upstream of the ﬂow meter. Using this quantity for characterizing the ﬂow has the advantage that its value, which is known for fully developed ﬂow, can be interpreted in physical terms, and that its measurement is nonintrusive, i.e., not causing an additional pressure drop. In the following we describe the characterization of the ﬂow by the fundamental disturbances as deﬁned in Chap. 2 of this book, the derivation of characteristic numbers by measuring the wall shear stress, and the establishment of the relationship for correcting the reading of the ﬂow meter. The validity of the correction procedure will be demonstrated by a number of experimental applications.

14.2 Characterization of the Disturbed Flow

225

14.2 Characterization of the Disturbed Flow By means of a theoretical approach Gersten and Papenfuss (Chaps. 2 and 3 in this book) show that the disturbed ﬂow in a pipe of circular cross section can be expressed as the composition of a basic ﬂow pattern and a set of superimposed secondary patterns that are regarded as fundamental disturbances. Each of these disturbances is deﬁned as a speciﬁc set of eigenfunctions of the problem and represents a speciﬁc ﬂow structure extending in the axial direction, e.g., pure swirl, a pair of two counterrotating vortices (“secondary ﬂow”, see Fig. 14.1a), a quadruple of vortices, etc.. The decay rates of the intensity of these ﬂow structures in the axial direction are different, e.g., the theory conﬁrms the experimentally known fact that the decay rate of pure swirl is the lowest among all fundamental disturbances. Our aim is to characterize the state of the disturbed ﬂow by measuring the wall shear stress along a circumference of the pipe and relating the measured result to the fundamental disturbance structures in Gersten and Papenfuss’ theory. 180°

90°

270°

a

0° 360°

90°

180°

T

Wtan

0

b

0°

270°

360°

azimuthal angle T

Fig. 14.1. (a) Pair of two counter-rotating vortices (“secondary ﬂow”) as caused by a 90◦ single bend and deﬁnition of the azimuthal angle with respect to the orientation of the bend. (b) Principal distribution of the tangential component of the wall shear stress downstream of the 2 × 90◦ out-of-plane bend: (1) contribution of swirl, (2) contribution of superimposed pair of counter-rotating vortices.

226

14 Correction of the Reading of a Flow Meter

The wall shear stress τ has two components, one in the axial and one in the azimuthal or tangential direction, τax and τtan , that will be measured as functions of the azimuthal angle θ (see Fig. 14.1a). For fully developed ﬂow one has τ ax = τ∞ = const. and τtan = 0. Figure 14.1b shows the principal distribution of τtan for swirl (in the ﬁgure designated as 1) plus the superimposed pair of counter-rotating vortices (designated as 2). For the purpose of ﬂow characterization we develop the non-dimensional components of the wall shear stress, T , into Fourier series: ∞ 2π 2π τax − τ∞ θ + bi ax sin i θ = a0 ax + ai ax cos i τ∞ 360 360 i=1 (14.1) ∞ 2π 2π τtan θ + bi tan sin i θ = a0 tan + ai tan cos i Ttan (θ) = τ∞ 360 360 i=1 Tax (θ) =

It can be shown that the Fourier coefﬁcients ai , bi can be associated to the eigensolutions of Gersten and Papenfuss’ theory and the respective fundamental disturbances or ﬂow structures. In particular, the following relationships exist: Fourier coefﬁcients

ﬂow structure

a0 ax a0 tan a1 ax , b1 ax a1 tan , b1 tan a2 ax , b2 ax a2 tan , b2 tan

ring vortex single axial vortex (pure swirl) pair of counter-rotating vortices source-sink pair quadruple of vortices source-sink quadruple

In Chap. 2 of this book Gersten and Papenfuss explain that disturbances associated with higher order coefﬁcients i > 2 decay very rapidly and can be disregarded for our further investigations. In the following we shall describe how the ﬁrst ten Fourier coefﬁcients are determined from the measured distributions T ax (θ) and Ttan (θ). Important for the envisaged correction procedure is that ten coefﬁcients are in practice sufﬁcient for an efﬁcient correction, i.e., that higher-order coefﬁcients can be neglected.

14.3 Experiments 14.3.1 Flow Facility The experiments are performed with air ﬂow in a pipe of circular cross section (diameter D = 100 mm) at pipe Reynolds numbers ranging from Re D = 5 × 104 to ReD = 2.5 × 105 . Higher values of the Reynolds number cannot be produced. This ﬂow facility is equipped with a device allowing continuous reference measurements of the volumetric ﬂow rate with an accuracy of ±0.25%. This device is based on a one-point measurement of the velocity proﬁle in fully developed ﬂow; for further

14.3 Experiments

227

details of the ﬂow facility see Wildemann (2000) and also Chaps. 4 and 8 in this book. Three different installations are used for disturbing the ﬂow in the pipe: a 90 ◦ single bend (radius of curvature of the centerline 1.5 D), a 2 × 90 ◦ out-of-plane double bend (for respective sketches see Chap. 4), and a gate valve, i.e., a circular plate, inserted through a slit from above as a gate, with its plane normal to the pipe axis, thus blocking off a certain percentage of the pipe cross section. 14.3.2 Measurement of Wall Shear Stress The wall shear stress τ is measured at various axial positions x/D downstream of the installations, ranging from 3 ≤ x/D ≤ 77, and in fully developed ﬂow. Sublayer fences (Dengel et al. (1987); Nitsche (1994)) inserted in the inner pipe wall serve for measuring τ . They are calibrated in fully developed ﬂow by making use of the balance between shear force and pressure force. Since it is necessary to determine two components of the wall shear stress (see (14.1)), two fences are used at each

a

1.5 2

3

W tan [N/m ]

2

Wax [N/m ]

44

2

0

1

-1.5

0

b

2

3

W tan [N/m ]

2

Wax [N/m ]

0.75

2

0 ax tan fully developed Fourier decomp.

1 41 D 0

0°

180°

-0.75 360°

T Fig. 14.2. Azimuthal distribution of wall shear stress downstream of the 90◦ single bend: Coordinate of the axial component, τax , is on the left, coordinate of the tangential component, τtan , on the right. Pipe Reynolds number ReD = 2.2 × 105 . The values of τax and τtan for fully developed ﬂow are indicated. (a): Axial distance from bend x/D = 3, Fourier decomposition of τax and τtan is included; List of Fourier coefﬁcients: a0 tan = −0.0206, a0 ax = 0.319, a1 tan = −0.199, a1 ax = 0.0128, b1 tan = −0.0080, b1 ax = 0.112, a2 tan = −0.0051, a2 ax = 0.0517, b2 tan = 0.0781, b2 ax = −0.0003. (b): x/D = 41.

14 Correction of the Reading of a Flow Meter

2

0

1

2

ax tan fully developed Fourier decomp.

41 D

3 2

0

1 0

-1.5 0.75 2

0

Wax [N/m ]

b

2

T

a

W tan [N/m ]

2

Wax [N/m ]

1.5

3D

3

0°

180°

W tan [N/m ]

228

-0.75 360°

T Fig. 14.3. Azimuthal distribution of wall shear stress downstream of the 2 × 90◦ out-of-plane double bend. The value of τax for fully developed ﬂow is indicated. (a): x/D = 3, ReD = 2.2 × 105 . Fourier decomposition of τax and τtan is included; List of Fourier coefﬁcients: a0 tan = −0.246, a0 ax = 0.122, a1 tan = 0.0654, a1 ax = 0.0838, b1 tan = 0.0974, b1 ax = −0.157, a2 tan = −0.0553, a2 ax = 0.0021, b2 tan = −0.0145, b2 ax = 0.0413. (b): x/D = 41.

measuring position that are oriented at ±45◦ with respect to the direction of the pipe axis. The two components, τax and τtan , are calculated with a formula describing the angular sensitivity of sublayer fences (Nitsche (1994)). Application of this formula, which was derived in the boundary layer ﬂow at a ﬂat plate, appears justiﬁed because the ratio of fence height (0.15 mm) and pipe diameter (100 mm) is very small. The azimuthal distribution τ (θ) is measured along a circumference of the pipe, i.e., for a range of the azimuthal angle θ from 0◦ to 360◦ . For this purpose the two sublayer fences are inserted in a ring whose inner surface is ﬂush with the inner pipe wall and which can be turned around the pipe axis. At stationary ﬂow conditions, τ is measured at angular intervals ∆θ = 5◦ , i.e., at 72 equally spaced angular positions θi for the whole circumference. Downstream of the 90◦ bend and the 2 × 90◦ out-of-plane double bend, these measurements are performed for the 5 downstream positions x/D = 3, 5, 11, 21, and 41, and for the 4 Reynolds numbers ReD /105 = 0.55, 1.3, 2.2, and 2.45. From this variety of measured data we show only four typical distributions of τax (θ) and τtan (θ) in Figs. 14.2 and 14.3. Also shown is the Fourier decomposition of τ ax and τtan ; the Fourier coefﬁcients are listed in the ﬁgure legend. The different scales for

14.3 Experiments

229

τax on the left and τtan on the right side of the diagrams should be noted. The distributions of τax and τtan are governed by the existence of the two counter-rotating vortices as shown in Fig. 14.1a,b. Therefore, τtan changes its sign downstream of the 90◦ bend (Fig. 14.2a), while this quantity remains always negative downstream of the double bend due to the superimposed negative swirl (Fig. 14.3a). At x/D = 41, τax approaches closely its value for fully developed ﬂow; this also applies to τ tan downstream of the single bend, i.e., τtan → 0 (Fig. 14.2b), while, for this position, τtan assumes an almost constant negative value downstream of the double bend (Fig. 14.3b), thus giving evidence of the low decay rate of the swirl generated by the double bend. Two sketches inserted in Figs. 14.2a and 14.3a, respectively, indicate the spatial orientation of the bends with respect to the ﬂow meter in the straight pipe section downstream of the installations. The four pressure taps for the Venturi and oriﬁce are located at the angular positions θ = 0◦ , 90◦ , 180◦ , and 270◦ ; the four pressure taps are connected by a hose such that a pressure value averaged along the circumference is measured. 14.3.3 Measurement of Discharge Coefﬁcients Three different ﬂow meters are exposed to the ﬂow disturbed by the three installations mentioned in Sect. 14.3.1: a Venturi according to ISO 5167-1 with an opening diameter ratio β = 0.7, and two oriﬁces according to ISO 5167-1 with β = 0.65 and β = 0.8. The value β = 0.8 is beyond the limits set by several norms for practical use of the oriﬁce; this value was chosen here only for demonstrating the method with a meter that is known to have an uncertainty in CD higher than normal. The differential pressure ∆p at the ﬂow meters is measured with a transducer (MKS Baratron 698) for which a value of the relative accuracy of ±0.1% is given by the producer, while the air density ρ, necessary for deriving the discharge coefﬁcient, is determined from a measurement of temperature and absolute pressure with a precision of ±0.15%. The discharge coefﬁcient CD of the meters, as deﬁned by equation (1) in Mattingly (1983), is then calculated from the measured data ∆p, ρ, the volumetric ﬂow rate Q provided by the reference measurement (see Sect. 14.3.1), and the geometry of the meter (β, throat area). A quantity (14.2) ∆CD = ((CD − CD∞ ) /CD∞ ) 100% in the literature often named “error shift” (Mattingly and Yeh (1991, 1994)), with CD∞ being the discharge coefﬁcient in fully developed ﬂow, is determined for all conﬁgurations of the installations, the indicated distances between ﬂow meter and installation, and the 4 values of the pipe Reynolds number realised in our experiments. The gate, one of the three installations, was set at three different positions, such that the pipe cross section, which remained free, was 98.7%, 88.2%, and 67.2%, respectively. This is equivalent to the position of the lower edge of the circular plate inserted from above at 0.1D, 0.3D, and 0.5D (see Fig. 14.4).

230

14 Correction of the Reading of a Flow Meter gate

D

Fig. 14.4. Circular ﬂat plate (gate valve) inserted from above into the pipe of diameter D as an installation.

As typical examples we show the values of ∆CD measured at various axial distances x/D between installation and meter for Venturi (Fig. 14.5) and the β = 0.65 oriﬁce (Fig. 14.6). Figures 14.5 and 14.6 indicate that the reading of both meters, Venturi and oriﬁce, is less affected by the disturbances caused by the 2 × 90 ◦ out-ofplane double bend, i.e., swirl, than by the disturbance due to the presence of a 90 ◦ single bend, i.e., “secondary ﬂow” due to the counter-rotating vortices. Particularly for the Venturi, the values of ∆CD are of the same order as the uncertainty in the measurement of the volumetric ﬂow rate. It is important to note that the values of 2.50

Reynolds number 120.000 105.000 180.000 160.000

'CD [%]

1.50

140.000 240.000

0.50

-0.50 0

10

20

30

40

50

60

70

80

90

100

-1.50 -2.50

x/D

2.50

Reynolds number 120.000 105.000 180.000 160.000

'CD [%]

1.50

140.000 240.000

0.50

-0.50 0

10

20

30

40

50

60

70

80

90

100

-1.50 -2.50

x/D

Fig. 14.5. Error shift ∆CD measured for the Venturi meter downstream of the 2 × 90◦ out-ofplane double bend (above) and the 90◦ single bend (below) as function of the axial distance x/D between installation and meter (horizontal scale).

14.4 Relationship Between Error Shift and Flow Disturbance 2.50

'CD [%]

1.50

Reynolds number 100.000 140.000

180.000

30

70

231

245.000

0.50 -0.50 0

10

20

40

50

60

80

90

100

-1.50 -2.50

x/D

2.50

Reynolds number 100.000 60.000

'CD [%]

1.50

140.000

245.000

180.000

0.50 -0.50 0

10

20

30

40

50

60

70

80

90

100

-1.50 -2.50

x/D

Fig. 14.6. Error shift ∆CD measured for the β = 0.65 oriﬁce meter downstream of the 2 × 90◦ out-of-plane double bend (above) and the 90◦ single bend (below) as function of the axial distance x/D between installation and meter (horizontal scale).

∆CD shown represent a systematic difference to reproducible values of CD . Here, their absolute numbers are not of any practical relevance, because they just serve to demonstrate the applicability and usefulness of the physical principle of correcting the reading of ﬂow meters.

14.4 Relationship Between Error Shift and Flow Disturbance According to our earlier assumption, a functional relationship exists, for a given meter, between the error shift ∆CD , measured for the various installations, and the ﬂow proﬁle whose disturbance is characterized by the Fourier coefﬁcients a i , bi (14.1). The Fourier coefﬁcients are functions of the distance x/D between installation and meter. This dependence expresses the decay rate of the particular disturbance in axial direction. A further parameter on which ai , bi depend is the pipe Reynolds number ReD , in our experiments a measure of the bulk velocity in the pipe or the volumetric ﬂow rate. As an example of the derivation of the Fourier coefﬁcients from the measured distributions Tax (θ), Ttan (θ), 'Fig. 14.7 shows the dependence of a0 tan and the combined coefﬁcient c1 tan = a21 tan + b21 tan on the axial distance x/D for a Reynolds number ReD = 1 × 105 ; here, a0 tan represents the swirl whose low decay rate is evident, while c1 tan characterizes the vortex pair as caused by the single

232

14 Correction of the Reading of a Flow Meter 0.30

a0 tan; c1 tan

0.20 0.10 0.00 -0.10

0

20

40

60

80

ao tan 2x90° bend ao tan 90° bend c1 tan 2x90° bend c1 tan 90° bend

-0.20 -0.30 -0.40

x/D

Fig. 14.7. Fourier coefﬁcients a0 , c1 = a21 + b21 determined at various distances (horizontal scale) downstream of the 2×90◦ out-of-plane double bend and 90◦ single bend; pipe Reynolds number ReD = 1 × 105 .

bend, and it is seen that this disturbance decays much faster in axial direction than the swirl. The postulated functional relationship between ∆CD , the Fourier coefﬁcients and the Reynolds number is established by means of an artiﬁcial neural network. For this purpose we use a “feed forward” type network and the respective “Matlab”, version 5.0, software. The network is “trained” according to the Levenberg-Marquardt approximation with a selected, limited set of combinations of Fourier coefﬁcients (up to order 2), related error shifts, and Reynolds number for each of the three meters. These data sets for the training are taken from a certain number of experiments performed with the three ﬂow meters and the two bends, but without using the data measured with the gate valve as an installation. The selection of the limited number of data from the complete data sets as input for the network was performed on the basis of a random choice program. The question for the further investigations is, how accurate can ∆CD be predicted from a measured distribution of Tax (θ), Ttan (θ) for those cases, that were not used in the training of the neural network. For details of the selection and the training of the network, see Wildemann (2000). The neural network is, of course, a substitution for a respective explicit relationship that is not available at this time; the network is easy to implement and it can be used without difﬁculties.

14.5 Results The “error shifts” predicted with the artiﬁcial neural network, ∆CD ANN , and measured in our experiments, ∆CD exp , are compared in Fig. 14.8 for the case of the Venturi meter. Again ∆CD ANN is determined via the Fourier coefﬁcients ai , bi from the measured distributions Tax (θ), Ttan (θ). The ﬁgure evidences which of the data were used for the training of the network and which are actually predicted. The data points shown include all positions x/D and Reynolds numbers as listed above. Ide-

14.5 Results

'CD ANN [%]

2

233

data used for training data used for testing

1 0 -1 -2

-2

-1

0

1

2

'CD exp [%] Fig. 14.8. Comparison of measured error shift, ∆CD exp , and error shift predicted by the artiﬁcial neural network, ∆CD ANN , for the Venturi meter.

ally, all data points should lie on the oblique straight line through the origin. The scatter which is due to the experimental inaccuracies and the approximations by the neural network indicates the accuracy (or inaccuracy) in the prediction of ∆C D . In order to quantify the precision of the prediciton we form the difference δ∆C = ∆CD exp − ∆CD ANN

(14.3)

whose dimension is %, since ∆CD is expressed in %, too, according to (14.2). Then, a probability density function (PDF) of δ∆C is determined (Fig. 14.9). From the pattern of the PDF and the area it includes one can derive that for 90% of the predictions the difference between the predicted and measured error shift is smaller than ±0.18%, while for 100% of the predicitons this difference is smaller than ±0.3%. From Figs. 14.8 and 14.9 it follows that the investigated Venturi meter, when being exposed to the pipe ﬂow disturbed by the present installations, measures the volumetric ﬂow rate with an inaccuracy of −1.0% ≤ ∆CD ≤ +0.7%, and that this uncertainty range is reduced to −0.3% ≤ ∆CD ≤ +0.3% with the device for correction described here. 5

PDF

4 3 2 1 -1

-0.8

-0.6

-0.4

-0.2

0

0

0.2

0.4

0.6

0.8

1

'CD exp - 'CD ANN Fig. 14.9. Probability density function (PDF) of the scatter of the data shown in Fig. 14.7.

234

14 Correction of the Reading of a Flow Meter

'CD ANN [%]

5

data used for training data used for testing

2.5 0 -2.5 -5 -5

-2.5

0

2.5

5

'CD exp [%] Fig. 14.10. Comparison of measured error shift, ∆CD exp , and error shift predicted by the artiﬁcial neural network, ∆CD ANN , for the β = 0.8 oriﬁce meter.

The respective results for the two oriﬁce meters are presented in Figs. 14.10 and 14.11. The measured error shifts are in most cases negative and their absolute range is larger than that for the Venturi. This conﬁrms the known fact that the Venturi is one of the most robust meters regarding the inﬂuence of ﬂow disturbances (see, e.g., Baker (2000)). The oriﬁce with β = 0.8 that is not deﬁned in the norms has error shifts larger than those for the oriﬁce with β = 0.65. From the PDF (not shown here) it follows that the uncertainty is reduced by the correction device to −0.6% ≤ ∆CD ≤ +0.6%, and for the oriﬁce with β = 0.65 that is designed according to the norms to −0.35% ≤ ∆CD ≤ +0.35%. We investigate next how the device and procedure for correction perform when being applied to the ﬂow disturbed by the installation whose data, i.e., the distributions Tax (θ) and Ttan (θ), were not used for “training” the artiﬁcial neural network. The installation is the gate sketched in Fig. 14.4 and used with the three different positions of the lower edge of the plate as described above. Also, the distance between installation and meter, x/D, and the Reynolds number are varied as indicated. These

'CD ANN [%]

5

data used for training data used for testing

2.5 0 -2.5 -5 -5

-2.5

0

2.5

5

'CD exp [%] Fig. 14.11. Comparison of measured error shift, ∆CD exp , and error shift predicted by the artiﬁcial neural network, ∆CD ANN , for the β = 0.65 oriﬁce meter.

14.5 Results

235

'CD ANN [%]

2 1 0

-1 -2

-2

-1

0

1

'CD exp [%]

2

Fig. 14.12. Comparison of measured error shift, ∆CD exp , and error shift predicted by the artiﬁcial neural network, ∆CD ANN , for the gate (installation shown in Fig. 14.4). No data measured with this installation were used for the training of the ANN.

experiments are restricted to the Venturi which is, among the three meters tested here, the least sensitive regarding disturbances. That is, the respective ∆C D values for the oriﬁces are expected to be larger than those for the Venturi, thus providing a higher signal amplitude for the correction. We conclude that, if the correction performs well with the Venturi, it should also work with those meters that are more severely affected by the disturbances than the Venturi. The result, presented in Fig. 14.12, is very satisfactory and can be taken as a proof of the applicability of the used principle that any disturbance caused by an installation is composed of a ﬁnite number of fundamental disturbances. In a practical application of this correction method it would be necessary to measure the wall shear stress (or a related quantity) with a set of stationary sensors along a circumference, as done here slightly upstream of the meter, and it is then of interest to minimise the number of sensors. For the results shown above we used the data of 72 measuring positions along the circumference for determining the distributions Tax (θ) and Ttan (θ). In order to provide information regarding the possible minimisa5

PDF

4

72 36 18 9

3 2 1

-1

-0.8

-0.6

-0.4

-0.2

0

6 0

0.2

0.4

'CD exp - 'CD ANN

0.6

0.8

1

Fig. 14.13. Probability density functions (PDF) as deﬁned and shown in Fig. 14.9 for different numbers of measuring positions used in determining the distributions Tax (θ) and Ttan (θ).

236

14 Correction of the Reading of a Flow Meter

tion of the number of measuring positions or sensors, we have also determined these distributions with data from a reduced number of measuring positions and used the new data for performing the correction procedure. The inﬂuence of the number of the measuring positions used, N, on the quality of predicting ∆CD is investigated with the PDFs as they have been explained above. With decreasing number N the amplitude of the PDF’s maximum decreases while the width increases, as shown in Fig. 14.13 for the Venturi meter. From this presentation one can easily derive how the accuracy limits within which the corrected error shift ∆CD is predicted vary with N.

14.6 Conclusions A method is described for correcting the reading of ﬂow meters that are exposed to pipe ﬂow disturbed due to the presence of an installation. The axial distance between installation and ﬂow meter can be minimised to a few pipe diameters; the smallest distance we investigated was 3 pipe diameters. In contrast to ﬂow conditioners that are also used for keeping this distance small, the presented method works without any additional pressure drop. The principle of the method is based on the theoretically supported assumption that the disturbances caused by installations can be classiﬁed into fundamental disturbances, here: speciﬁc ﬂow structures, whose linear superposition allows characterization of the disturbed pipe ﬂow (cf. Chaps. 2 and 3). These fundamental disturbances are detected with a device, located slightly upstream of the meter, that, in our case, measures the azimuthal distribution of the two components of the wall shear stress. We postulate that a functional relationship exists between the set of detected characteristic disturbances and the “error shift” of the ﬂow meter. This relationship is established by means of an artiﬁcial neural network, and the correction is successfully demonstrated, even for an installation that is, in principle, “unknown” to the network. We have taken this result as proof that the characterization of the disturbed ﬂow by means of the fundamental disturbances is realistic. An advantage of characterizing the disturbed ﬂow velocity proﬁles in the described way is that the fundamental disturbances used have a physical signiﬁcance, namely characteristic ﬂow structures, which become evident in the measured distributions of the wall shear stress. In developing and deriving the correction procedure, we did not make use of the physical signiﬁcance of the quantitatively determined wall shear stress. The measured distributions of τ , shown in Figs. 14.2 and 14.3, are helpful for the physical interpretation of the correction procedure. But any signal giving quantitative evidence of the deviation from the fully developed state along the circumference could be used for the same purpose. This means that, for a practical application of the correction principle, the sublayer fences can be replaced by other sensors that must only fulﬁll the condition of having a reproducible sensitivity regarding changes of the wall shear stress. Inexpensive semiconductor sensors of the required type are available that can be considered for a practical realisation. Technical details remain to be investigated

References

237

for an adaptation of the correction principle to use in practice, e.g., minimisation of the number of sensors along the circumference, optimisation of the neural network and its training, replacing the neural network procedure with an algebraic functional relationship, further testing with a variety of different installations as well as different ﬂow meters. The work presented here can be considered to deliver the physical basis for such developments, see Gersten et al. (2001). Problems of compatibility of the method with rules set by technical norms should only become relevant when investigations of the mentioned technical details have arrived at satisfactory solutions.

References Baker, RC (2000) Flow Measurement Handbook. Cambridge University Press Dengel P, Fernholz HH, Hess M (1987) Skin-friction measurements in two- and threedimensional highly turbulent ﬂows with separation. In: Comte-Bellot G, Mathieu J (Eds): Advances in Turbulence, Springer-Verlag, Berlin/Heidelberg, pp. 470–479 Gersten K, Klika M (1998) The decay of three-dimensional deviations from the fully developed state in laminar pipe ﬂow. In: Rath HJ, Egbers C (Eds.): Advances in Fluid Mechanics and Turbomachinery, Springer-Verlag, Berlin/Heidelberg, pp. 17–28 Gersten K, Merzkirch W, Wildemann C (2001) Verfahren und Vorrichtung zur Korrektur fehlerhafter Messwerte von Durchﬂussmessgeräten infolge gestörter Zuströmung. Patent No. 197 24 116 Laws EM (1990) Flow conditioning - A new development. Flow Meas Instrum 1:165–170 Mattingly GE (1983) Volume ﬂow measurements. In: Goldstein RJ (Ed.): Fluid Mechanics Measurements, Hemisphere Publ Corp, Washington DC, pp. 245–306 Mattingly GE, Yeh TT (1991) Effects of pipe elbows and tube bundles on selected types of ﬂow meters. Flow Meas Instrum 2:4–13 Mattingly GE, Yeh TT (1994) Pipeﬂow downstream of a reducer and its effect on ﬂowmeters. Flow Meas Instrum 5:181–187 Mickan B, Wendt G, Kramer R, Dopheide D (1996) Systematic investigation of pipe ﬂows and installation effects using laser Doppler anemometry. Part II: The effect of disturbed ﬂow proﬁles on turbine gas meters - A describing empirical model. Flow Meas Instrum 7:151–160 Nitsche W (1994) Strömungsmesstechnik. Springer-Verlag, Berlin Reader-Harris MJ, Sattary JA, Spearman EP (1995) The oriﬁce plate discharge coefﬁcient equation - Further work. Flow Meas Instrum 6:101–114 Wildemann C (2000) Ein System zur automatischen Korrektur der Messabweichung von Durchﬂussmessgeräten bei gestörter Anströmung. Doctor thesis, Universität Essen. See also: Fortschritt-Berichte VDI, Reihe 8, Nr. 868 Xiong W, Kalkühler K, Merzkirch W (2003) Velocity and turbulence measurements downstream of ﬂow conditioners. Flow Meas Instrum 14:249–260

15 How to Correct the Error Shift of an Ultrasonic Flow Meter Downstream of Installations Carsten Ruppel, Franz Peters Institut für Strömungslehre, Universität Essen, 45117 Essen, Germany

A commercial ultrasonic ﬂow meter in combination with a new probe ﬂange is subjected to the disturbed ﬂow downstream of a 90 ◦ bend and a double bend, respectively. The probe ﬂange determines the ﬂow structure by swirl-angle measurements. Simultaneously, the error shift of the meter is recorded. A mathematical link is established that relates the ﬂow structure to the error shift. With this link the error shift can be predicted for any position of the meter, which means that the meter can be corrected. It is shown that not only the mean error shift, but also the circumferential error shift of the meter, can be signiﬁcantly reduced.

15.1 Introduction The problem of the error shift correction of a ﬂow meter mounted downstream of an installation has been addressed at length in Chap. 14 saving a discussion at this point. However, as the present work builds up on the key results of that chapter we want to recall these as a starting point. It was shown that the disturbances induced by the installations (secondary ﬂows) can be experimentally identiﬁed in terms of the wall shear stress distribution. The modeling of these distributions using Fourier series proved feasible as had been postulated in Chaps. 2 and 3. And considerable corrections of a Venturi and of two oriﬁces could be veriﬁed when the distributions were related to the error shifts using a neural network transformation. While these results form the scientiﬁc basis of any correction procedure they do not yet concern any application aspects. At present we want to deal with these centered around a commercial ultrasonic ﬂow meter. The sublayer fence for the wall shear stress measurement as used in Chap. 14 is out of the question for practical application. It is too sensitive to clogging and contact. The encountered low pressure differences require high resolution, i.e., expensive transducers in combination with long calibration times. An array of these sensors along the pipes cirumference, as would be needed for on-line measurements, is unrealistic. In consequence we have developed a new differential pressure probe to meet the following requirements

240

-

15 How to Correct an Ultrasonic Flow Meter

rugged design with high signal to noise ratio inexpensive pressure transducers fast signal generation as a prerequisite for on-line correction capability for array measurements simple with respect to implementation into real piping systems (ﬂange design)

It no longer measures the wall shear stress but the swirl angle close to the wall, which turns out to be the better choice. Concept and performance of the single sensor has been investigated ﬁrst (Peters and Ruppel, 2004). Later (Ruppel and Peters, 2004) an array of sensors was integrated into a ﬂange. We are using this ﬂange in combination with a commercial meter with the focus on the correction of the meter. No new technical developement will be reported. As a second accomodation to application we do abandon the neuronal network for two reasons. The inherent mathematical transformation is not transparent such that the inﬂuence of parameters cannot be traced. And secondly, for an on-line correction the involved calculation is too complex. A simple expression composed of power functions replaces the neuronal network with success. At present the pressure probe array is used downstream the installations 90 ◦ bend and double bend in combination with a commercial ultrasonic ﬂow meter. The meter under investigation is of the transit-time type (Baker, 2000) which makes use of the fact that sound is carried by the ﬂuid at an effective speed with respect to the wall which is either the difference or the sum of its own speed and the ﬂuid velocity. The sound is directed across the streaming ﬂuid as a beam such that velocity information is collected along a line (line sensor) in terms of time shift or frequency without being able to reconstruct the velocity proﬁle. Therefore, there is a principle ambiguity in determining the mean transport velocity when the proﬁle is not known. Various designs try to circumvent this problem by multiple paths arrangements which penetrate the stream in more than one direction. Some designs claim (Baker, 2000; Holden and Peters, 1991) to attain measuring uncertainties below 1%. These remarkable achievements together with the crucial beneﬁts of an unblocked cross section and applicability to both gases and liquids have led to a great success with a variety of models on the market. It was important to us to select a widespread commercial meter to be sure that the desired correction is practically relevant. As to the detailled analysis of the meter’s error shift the reader is referred to a previous publication (Ruppel and Peters, 2004). The key issue of this work is on-line correction which means that the probe ﬂange is always in place together with the meter. We are neither analyzing the ﬂow nor the meter separately. The combination of both instruments is to be seen as a new instrument.

15.2 Experimental Setup 15.2.1 Test Rig The experiments were carried out in our multifunctional ﬂow rate test rig a top view of which is sketched in Fig.15.1. A speed controlled blower circulates room

15.2 Experimental Setup reference section

d= 100mm

venturi meter

reducer

ultrasonic flow meter 2D

installation: 90°-bend

A

A probe flange

180°

4=0°

90°

Section A-A

270°

installation: double-bend

13.3 D

straight pipe

36 d

blower

m

test section distance

D= 150mm

20 d

241

Akashi flow straightener

m

m

Fig. 15.1. Top view of test rig.

air through the test and reference sections made of precision piping of 150 mm and 100 mm inner diameter, respectively. The test section is segmented and assembled by ﬂanges to allow variable conﬁgurations of pipe lengths and installations including the ultrasonic meter and the pressure probe ﬂange. The reference section incorporates a Venturi at 20d upstream of the blower (d = 100 mm) and 36d downstream of a constriction which reduces the wider pipe (D = 150 mm) to the smaller one. The Venturi was calibrated against another reference based on developed pipe ﬂow which can be placed optionally on the downstream side of the blower (Wildemann et al. , 2002). By means of pressure and temperature measurements in the Venturi the Venturi differential pressure reading is converted into mass ﬂux. The mass ﬂux reproducibility is better than ±0, 2% over all tested conﬁgurations in the test section upstream of the constriction. This is an important feature because we are looking for error shifts. The absolute measuring uncertainty is less important. 15.2.2 Probe Flange Figure 15.2 provides a photograhic picture of the probe ﬂange looking downstream. The eight probes protrude radially from the wall into the ﬂow for a few millimeters. They are wedge-shaped with the edge pointing straight upstream. Each probe features two pressure taps symmetrically located on the two sides of the wedge. The two pressures are individually measured against the static pipe pressure amounting to sixteen individual low cost pressure transducers (PXLA 12x5 GN by Sensortechnics). The information drawn from the probe measurement is the ﬂow angle θ close to the wall vs circumferential angle Θ forming a clear picture of the vortex formation behind installations. Eight probes provide sufﬁcient information for the Fourier

242

15 How to Correct an Ultrasonic Flow Meter

Fig. 15.2. Photography of probe ﬂange.

modeling of the ﬂow angle distribution. All this was shown in Ruppel (2004).The pressure signals were digitized by the A/D board DAQ Pad 6020E of National Instruments at a sampling frequency of 1000 Hz averaging over 90 s. All data processing was carried out by LabView 6i of National Instruments. 15.2.3 Meter A commercially available ultrasonic ﬂow meter for gas was used. It is layed out for the pipe diameter of 150 mm with ﬂange attachements according to DIN 2633. Its design is based on two V paths. The beams are not directed radially and not even in a radial plane. They penetrate a measuring volume in an optimized fashion the background of which is a corporate secret. The meter is sold for the volume ﬂow rate range 14 − −1600 m3/h corresponding to the mass ﬂux range 5 − −533 g/s under our test conditions. The reading of the meter is in m3 /s. Like with the Venturi we convert to mass ﬂux by measuring pressure and temperature continuously 1D downstream of the meter. The output signal of the meter is a current between 4 mA to 20 mA proportional to the ﬂow rate which is processed by a PCI 6024 by National Instruments. Integration time is 90 s at a sampling frequency of 1000 Hz. Figure 15.3 provides the characteristic curve of the meter with the ﬂange in place in a developed pipe ﬂow(50D of straight pipe). A perfect linearity between the output current and the ﬂow rate is observed. From three different test series (see legend) it was calculated that the ﬂow rate can be predicted from the measured current with an uncertainty of less than ±0.15%.

15.2 Experimental Setup

, (mA)

20 18 16 14 12 10 8 6 4 2 0

243

1. test series 2. test series 3. test series mean value

0

100

200 300 m ref / U (dm³/s)

400

500

Fig. 15.3. Characteristic curve of the ultrasonic meter with probe ﬂange for developed pipe ﬂow.

15.2.4 Orientation In all of the present experiments the probe ﬂange and the ultrasonic meter were assembled as a ﬁxed unit. Figure 15.4 gives the relative angular position of probes and meter and also the distance between the two which is 2D. In this investigation the unit takes a number of different positions relative to the upstream installations in terms of distance and 12 different positions in terms of angle. (The manufacturer recommends a distance of 3D to the next upstream installation to keep accuracy speciﬁcations. He stipulates no angular orientation of the meter.) As shown in Fig. 15.1 the distance is measured from the meters center plane and the angle is given in terms of ΘUZ . The latter measures from Θ = 0 as deﬁned by the section A-A in Fig. 15.1 up to the meters circumferential reference line pointed out in Fig. 15.4. According to the top view of Fig. 15.1 Θ = 0 refers to the top position while 270 ◦ lies on the inside of the bend.

ultrasonic meter circumferential reference line for 4UZ

distance reference line

2D

probe flange

ultrasonic probe pressure probe pipe

Fig. 15.4. Relative positioning of meter and probe ﬂange.

244

15 How to Correct an Ultrasonic Flow Meter

15.2.5 Installations The error shifts caused by two typical installations, the 90◦ -bend and the double bend, are studied. Their radius to diameter ratio is 1.5 following DIN 2605. The general ﬂow patterns induced by these bends is as follows (more details in Peters and Ruppel (2004)). When axial pipe ﬂow (free of swirl) enters the 90 ◦ bend a counterrotating vortex pair appears at the exit that was already described by Dean (1928). When the bend merges into a straight pipe this vortex pair looses strength along the pipe, yet it maintains its circumferential position. The entire ﬂow disturbance induced by the bend vanishes after roughly 50D and developed pipe ﬂow prevails. This was shown experimentally by Schlüter and Merzkirch (1996) and also by Kalkühler (1998). When axial pipe ﬂow (free of swirl) enters the double bend CFD calculations (Hilgenstock and Ernst, 1996) and measurements (Xiong et al., 2003) show an excentric main vortex with an embedded smaller one. The smaller one disappears rapidly along the pipe while the excentric pattern of the bigger one rotates about the pipe axis with the ﬂow. At a ﬁxed axial position in the pipe the pattern stays put. Decay of the main vortex is rather slow with developed pipe ﬂow restored only after 100D (Xiong et al., 2003; Schlüter and Merzkirch, 1996). In a real piping system the bend will be fed from an upstream pipe or just another bend or installation of some sort which means that the inlet conditions to the bend creating the vortex ﬂow is a disturbed ﬂow itself which is most likely not stable. Any numerical calculation on the bend ﬂow or any experiment has, in principle, to take into account the individual inlet conditions which they normally don’t. Because of lacking lab space we cannot extend the inlet pipe to a length that provides undisturbed developed ﬂow. Yet, in order to secure stable inlet conditions, irrespective of the radial proﬁle, we found the Akashi ﬂow straigthener (Akashi et al., 1978) to work well (Fig. 15.1) in contrast to an inlet nozzle which suggests smooth ﬂow but enhances swirl and instabilities. This aspect of the upstream stability in context with meter correction has not been addressed before by other investigators although it is a very important prerequisite to success. The correction procedure relies on a detectable and reproducible ﬂow structure which can only be achieved by stable upstream conditions. It is a widespread misconception that the bend ﬂow itself is instable and one has to live with it. Therefore, for good correction results measures have to be taken upstream of the installation, not downstream. Generally, the quality of the intelligent correction depends mainly on the uniqueness of the upstream ﬂow.

15.3 Experiments The ﬁrst experiments were conducted with the 90◦ bend at 7 distances between 4D and 20D. The probe ﬂange records the swirl angle θ versus the circumferential angle Θ appearing in Fig. 15.5a for the 4D and in Fig. 15.5b for the 12D distance. It is an eye-catching feature that the closer distance has the higher amplitudes with

15.3 Experiments

a)

245

b) 20

20

15

4D

10

swirl angle T (°)

swirl angle T (°)

15 5 0 -5

-10

12D

10 5 0 -5

-10

-15

-15

-20

-20 0

45

90 135 180 225 270 315 circumferential position 4 °)

360

0

45

90 135 180 225 270 315 circumferential position 4 °)

360

Fig. 15.5. Swirl angle vs circumferential angle. Symbols for measured points (◦) ΘUZ = 0◦ , () ΘUZ = 30◦ , () ΘUZ = 60◦ and lines for Fourier series according to (15.1) (—–) ΘUZ = 0◦ , (- - -) ΘUZ = 30◦ , (· · ·) ΘUZ = 60◦ . Installation: single bend; Re=200,000.

swirl angles up to ±15◦ . This reﬂects the amplitude decay along the pipe that has been observed before (Ruppel, 2004). Also notable is that the swirl angle proﬁle stays put from distance to distance with zeros at 90◦ and 270◦ and extreme values symmetric about the second zero. This agrees perfectly with the vortex pair found behind the bend (see Chap. 14) with the zeros corresponding to opposite stagnation points where the vortices part at 90◦ and merge at 270◦ . In this regard an angular turn of the unit should make no difference. In order to see that 12 different positions were tested. Three of them are plotted by different symbols as identiﬁed in the legend. The symbols line up nicely on a smooth common curve which is not plotted. The curves that are plotted are approximations of the data to be explained next. As pointed out earlier the ﬁrst step of the intelligent correction procedure is the modeling of the upstream ﬂow which is currently characterized by the swirl angle distribution θ(Θ). Now, as this distribution oscillates about a mean value and as it returns in itself after 2π a Fourier series is appropriate for modeling (see Chap. 14). Fθ (Θ) = θ(Θ) = a0 +

3

ai · sin (iΘ) + bi · cos (iΘ)

(15.1)

i=1

The series contains 7 coefﬁcients corresponding to the 8 data points provided by the probe ﬂange. With the data already acquired by LabView a best ﬁt calculation of the coefﬁcients was done in one go. Despite some residual sinusoidal oscillations the resulting curves represent the data quite well (a better ﬁt would ask for more coefﬁcients and more probes). The Reynolds number effect is marginal. For the smaller amplitude at the larger distance the deviations shrink correspondingly and the curves seem to coincide. Along with the swirl angle the error shift ∆FD ∆FD [%] =

˙ UZ m ˙ ref − m 100 m ˙ ref

(15.2)

246

15 How to Correct an Ultrasonic Flow Meter

a)

b) 4 2

2

1 0

-1 -3

1 0

-1

Re 250,000 Re 200,000 Re 150,000 mean value

-2

12D

3 'F D (%)

'F D (%)

4

4D

3

Re 250,000 Re 200,000 Re 150,000 mean value

-2 -3

-4

-4 0

45

90 135 180 225 270 315 circumferential position 4 UZ (°)

360

0

45

90 135 180 225 270 315 circumferential position 4 UZ (°)

360

Fig. 15.6. Error shift of the meter vs circumferential angle. Installation: single bend.

of the ultrasonic meter was recorded. It is deﬁned as the mass ﬂux deviation of the meter relative to the reference in %. In Figs. 15.6a and 15.6b we ﬁnd the results plotted versus the circumferential position of the unit in terms of ΘUZ giving the turn of the meter on the Θ scale (see Fig. 15.1 section A-A). At 4D a pronounced error shift proﬁle appears, independent of Reynolds number. Notable three percent are found at two distinct positions qualifying the meter as being generally sensitive to the identiﬁed ﬂow structure. However, with the availability of this curve the meter may be set to speciﬁc positions around 270◦ where the error is avoided to the greatest possible extent. Figures 15.7a and 15.7b display the equivalent results for the double bend installation. Strikingly, the 4D proﬁle is similar to the previous one, it just appears shifted downwards to negative swirl angles and along the abscissa. Details of this ﬂow have not yet been explored sufﬁciently. The general trend, however, can be explained by an additional rotationally symmetric vortex supposedly induced by the second bend. This vortex superposed to the vortex pair of the single bend obviously reduces all

a)

b) 10

10 5

4D

0

swirl angle T °)

swirl angle T (°)

5 -5

-10

11D

0 -5

-10

-15

-15

-20

-20

-25

-25

-30

-30 0

45

90 135 180 225 270 315 circumferential position 4 (°)

360

0

45

90 135 180 225 270 315 circumferential position 4 °)

360

Fig. 15.7. Swirl angle vs circumferential angle. Symbols for measured points (◦) ΘUZ = 0◦ , () ΘUZ = 30◦ , () ΘUZ = 60◦ and lines for Fourier series according to (15.1) (—–) ΘUZ = 0◦ , (- - -) ΘUZ = 30◦ , (· · ·) ΘUZ = 60◦ . Installation: double bend; Re=200,000.

15.4 Correction

a)

247

b) 4 2

2

1 0

-1 -3

1 0

-1

Re 250,000 Re 200,000 Re 150,000 mean value

-2

11D

3 'F D (%)

'F D (%)

4

4D

3

Re 250,000 Re 200,000 Re 150,000 mean value

-2 -3

-4

-4 0

45

90 135 180 225 270 315 circumferential position 4 UZ (°)

360

0

45

90 135 180 225 270 315 circumferential position 4 UZ (°)

360

Fig. 15.8. Error shift of the meter vs circumferential angle. Installation: double bend.

swirl angles along the circumference likewise shifting the curve downwards (direction is due to sign convention of course). At the same time the entire vortex pattern starts to revolve about the pipes axis as the pattern moves along. This comes forward when comparing the positions of the 4D and the 11D proﬁles with respect to Θ. The amplitude decays with traveling distance but not as strong as in the 90 ◦ case, an observation already made by Kalkühler (1998) and Schlüter and Merzkirch (1996) (see also Chap. 4). The error shifts come out accordingly. With the smaller amplitude decay a considerable sensitivity to angular position remains at 11D which had vanished in the single bend case.

15.4 Correction Now, that we know how to identify the ﬂow downstream of bend and double bend by Fourier coefﬁcients and how to monitor the corresponding error shifts of the ultrasonic ﬂow meter a mathematical link is required between the Fourier coefﬁcients and the error shifts. This link is to be established on basis of the experimental data obtained for bend and double bend. In its ﬁnal form it should be capable of predicting the measured error shifts from the measured Fourier coefﬁcients. When this has been achieved the elimination of the error shifts (the correction) can be performed for any installation situation in terms of distance and circumferential angle that was not among the ones used for setting up the link. Eventually, it will not be necessary to know which bend of the two is actually installed. The Fourier coefﬁcients provide all the input needed. Detailing all the necessary equations would go beyond the scope of this paper. The reader is referred to the thesis of Ruppel (2004). Here we resort to explaining the general concept with the help of Fig. 15.9. As already shown the error shift of the ultrasonic ﬂow meter depends on circumferential position and distance while Reynolds number effects are marginal. It makes sense, therefore, to compose the error shift of a mean value ∆Fmean that averages

248

15 How to Correct an Ultrasonic Flow Meter

the circumferential values and the circumferential value ∆Fcirc. itself offset by the mean such that it appears as an oscillation about zero. ∆F = ∆Fmean + ∆Fcirc.

(15.3)

The ﬁrst depends only on distance and the second on angular position and distance. The bottom left plots in Fig. 15.9 show the mean error shifts at various tested distances including the 4D and 12D (11D) positions exempliﬁed before. To the right there are the circumferential contributions displayed for only two positions each to maintain clarity. The Fourier coefﬁcients a0 , ai and bi (i=1,2,3) of (15.1) contain the full information on the swirl angle distribution. Particular combinations of these coefﬁcients provide measures for the amplitude and the phase of the signal (see Fourier analysis literature). The amplitude is associated with (15.4) c0 = a20 , ci = a2i + b2i The evaluated c0 and ci are plotted vs distance in the top left graphs with the decay clearly visable. The points for each c0 and ci correspond to different angular positions of the unit and should lie on a common curve. The residual scatter is due to the limited number of probes (see Fig. 15.5, 15.7). The coefﬁcients di and ei di =

bi ai , ei = ci ci

(15.5)

indicate the angular position of the signal (phase). The top right plots present in terms of these coefﬁcients what was said before: in the 90 ◦ case the position of the swirl angle distribution stays put from one distance to the next and in the double bend case the whole pattern rotates about the pipes axis. For clarity only the e 1 and d2 are plotted at two distances. In principle the 90◦ plot contains 36 curves and the double bend plot 42 curves. We now link the upper and lower parts of Fig. 15.9. The modeling of ∆Fmean relies on the cj (j = 0, 1, 2, 3) as both of them are independent of angle. The ansatz for ∆Fmean is based on a series of power functions of the form factorj cj exponentj

(15.6)

in which factor and exponent are determined by data ﬁtting such that ∆Fmean in the bottom plots is predictable by the cj of the upper plots. The ansatz for ∆Fcirc. is similar. The power functions are now extended by di and ei to account for the angular dependence. factorji cj exponentji di

and

factorji cj exponentji ei

(15.7)

Again factors and exponents come from data ﬁtting. When the two series of power functions are completed the error shift of the ultrasonic ﬂow meter can be

15.4 Correction 1.5

16

90°-bend

14 12

c0

10

c1 c2 c3

8 6

249

90°-bend

1 0.5 0 -0.5

4 -1

2

e1-4D

-1.5

0 0

2

4

6

8

10 12 14 distance (D )

16

18

20

22

0

24

16

45

double-bend c1 c2 c3

10 8 6

d2-12D 315

360

double-bend

1

c0

12

e1-12D

90 135 180 225 270 circumferential position 4 (°)

1.5

14

d2-4D

0.5 0 -0.5

4 -1

2 0

e1-4D

-1.5 0

2

4

6

8

10 12 14 distance (D )

16

18

20

22

0

24

4

90 135 180 225 270 circumferential position 4 (°)

d2-11D 315

360

4

90°-bend

3

90°-bend

3 2

2 'F circ. (%)

' F mean (%)

e1-11D

series of power functions (15.7)

series of power functions (15.6)

1 0

1

4D 12D

0

-1

-1

-2

-2

-3

-3 -4

-4

0

2

4

6

8

0

10 12 14 16 18 20 22 24 distance (D )

4

45

4

double-bend

3

90 135 180 225 270 circumferential position 4 (°)

315

360

315

360

double-bend

3

2

2 'F circ. (%)

' F mean (%)

45

d2-4D

1 0

-1

1 0

-1

-2

-2

-3

-3

4D 11D

-4

-4

0

2

4

6

8

10 12 14 16 18 20 22 24 distance (D )

0

45

90 135 180 225 270 circumferential position 4 (°)

Fig. 15.9. Correction procedure with Fourier coefﬁcients (top) and error shifts (bottom) to be predicted by the power functions.

250

15 How to Correct an Ultrasonic Flow Meter

predicted, which means it can be corrected. At any installation distance or angle just the Fourier coefﬁcients have to be determined by the probe ﬂange to calculate the correction. The completed correction procedure is bound to the respective unit. An exchange or modiﬁcation of the unit requires a new cycle through the whole procedure.

15.5 Results Figure 15.10 shows the results of correction for the bend (top) and the double bend (bottom) obtained for a set of data measured at the Reynolds number 200, 000. Again the distances 4D and 12D (11D) are selected for demonstration. ∆F is the measured error shift (compare Figs. 15.6, 15.8) while K represents the predicted error shift. R stands for the residual error which is the deviation between the two. The agreement is not perfect but most importantly the residual error is much smaller than the initial one.

b)

4

4

3

3

2

2

'F , K , R (%)

'F , K , R (%)

a)

1 0

-1 -3 -4 0

45

'F K R

-2 -3 -4

90 135 180 225 270 315 circumferential position4UZ q

360

c)

0

45

90 135 180 225 270 315 circumferential position4 UZ q

360

d)

4

4

3

3

2

2

'F , K , R (%)

'F , K , R (%)

0

-1

'F K R

-2

1

1 0

1 0

-1

-1

'F K R

-2 -3 -4 0

45

90 135 180 225 270 315 circumferential position4 UZ q

'F K R

-2 -3 -4 360

0

45

90 135 180 225 270 315 circumferential position4 UZ q

360

Fig. 15.10. Measured error shifts ∆F , predicted error shift K and residual error R vs circumferential position of the meter ΘUZ . Mounting positions: (a) 4D downstream the single bend, (b) 12D downstream the single bend, (c) 4D downstream the double bend, (d) 11D downstream the double bend. Re=200,000.

K (%)

15.5 Results

251

5 4 3 2 1 0 -1 -2 -3 -4 -5 -5

-4

-3

-2

-1 0 1 'F (%)

2

3

4

5

Fig. 15.11. Predicted error shift vs measured error shift. All investigated mounting positions.

Figure 15.11 plots K versus ∆F for all data that where taken at distances from 4D to 21D at angular positions between 0◦ and 360◦ and different Reynolds numbers from 150, 000 to 250, 000. Out of these data about a third was used to establish the mathematical link. If the correction was perfect all points would lie on the diagonal line. Therefore, the scatter about this line visualizes the achieved degree of correction. In order to quantify the distribution of points about the diagonal a probability density function (PDF) of R is provided in Fig. 15.12. It shows how many points in % on the ordinate fall into error classes on the abscissa. Noteable 31% of R fall into the ±0.1% class which means that almost a third of all measurements correspond to an error shift of less than ±0.1%. The curve drops off rapidly within the 1% class. The largest R appear between 0.75% and −0.79% at vanishing probabilities. For comparison the original distribution of ∆F data is given. They peak at 1.4% and lie between +3.34% and −2.09%. In order to characterize the correction by an overall benchmark one may take the average of all absolute ∆F which amounts to 1.26% and compare it with the average of all absolute R which is 0.21%. Then the error shift is reduced by 83%. 50 45 (%) 40 35 30 25 20 15 10 5 0 -5

R

'F

-4

-3

-2

-1

0 1 class (%)

2

3

4

5

Fig. 15.12. Frequency of occurance vs class of error shift for residual error R and measured error shift ∆F .

252

15 How to Correct an Ultrasonic Flow Meter

15.6 Conclusion After the feasibility of the error shift correction of ﬂow meters had been demonstrated by the work in Chap. 14 we tackled the problem of designing an instrument capable of identifying the ﬂow structure downstream of installations under realistic application conditions. A rugged ﬂange design featuring eight differential pressure probes emerged which measures the swirl angle close to the pipe wall within ±20 ◦ . The swirl angle modeled by Fourier series identiﬁes the vortex ﬂow structure successfully. A representative commercial ultrasonic ﬂow meter was chosen to be corrected for error shifts downstream of bend and double bend, respectively. The ﬂange was combined with the meter to form a unit mountable at user-deﬁned distances and circumferential positions. A straightforward mathematical link between ﬂow structure and error shift was established with great success. Thus, as a general result, we achieved to develope a correction procedure applicable to realistic conditions. As to the tested ultrasonic ﬂow meter it turned out that the meter should be corrected not only for distance but also for circumferential position. By the correction performed the reading of the meter could be improved by 83%. An important fact that we have adressed before and want to stress again is that the success of the correction depends heavily on the stability of the ﬂow. In case of bend ﬂow a simple ﬂow straightener helps (upstream the bend!). To conclude, we think that there is still potential for further improvements because information beyond the swirl, contained in the axial velocity proﬁle, has not yet been exploited.

References Akashi K, Watanabe H, Koga K (1978) Flow rate measurement in pipe line with many bends. Mitsubishi Heavy Ind 15:87–96 Baker RC (2000) Flow Measurement Handbook. Cambridge University Press Dean WR (1928) The streamline motion of ﬂuid in a curved pipe. Phil.Mag.Ser.7(5):673–695 Hilgenstock A, Ernst R (1996) Analysis of installation effects by means of computational ﬂuid dynamics-CFD vs experiments?. Flow Meas.Instrum. 7:161–171 Holden JL, Peters RJW (1991) Practical experiences using ultrasonic ﬂow meters on high pressure gas. Flow Meas. Instrum. 2:69–73 Kalkühler K (1998) Experimente zur Entwicklung der Geschwindigkeitsproﬁle und Turbulenzgrößen hinter verschiedenen Gleichrichtern. Dissertation, Universität Essen; also published as VDI Fortschritt-Bericht, Reihe 7, Nr. 339, 1998, VDI-Verlag, Düsseldorf Peters F, Ruppel C (2004) A pressure probe for the detection of the ﬂow direction close to walls. Case study: ﬂow through a bend. Exp. in Fluids 36:813–818 Ruppel C (2004) Die intelligente Korrektur der Fehlerverschiebung eines Ultraschall-Zählers bei gestörter Zuströmung. Dissertation Universität Duisburg-Essen. See also: ShakerVerlag Aachen ISBN 3-8322-2414-9 Ruppel C, Peters F (2004) Effects of upstream installations on the reading of an ultrasonic ﬂowmeter. Flow Meas. Instrum. 15:167–177

References

253

Schlüter T, Merzkirch W (1996) PIV measurements of the time-averaged ﬂow velocity downstream of ﬂow conditioners in a pipeline. Flow Meas. Instrum. 7:173–179 Wildemann C, Merzkirch W, Gersten K (2002) A universal, nonintrusive method for correcting the reading of a ﬂow meter in pipe ﬂow disturbed by installation effects. J Fluids Eng 124:650–656 Xiong W, Kalkühler K, Merzkirch W (2003) Velocity and turbulence measurements downstream of ﬂow conditioners. Flow Meas. Instrum. 14:249–260

Index

aerodynamic torque 169, 173, 177 artiﬁcial neural network 232 bluff body 216 circular form 105 pressure loss 107 T-shape 103 threaded rod 106 triangular 99, 102 vortex shedding (see also: Kármán vortex street) 90, 95, 99, 102, 211, 221 bulk velocity see pipe ﬂow coherent structure 130 correction method 235, 248, 250 cross-correlation function 80, 85, 112–114, 132, 133, 141 defect law 30 discharge coefﬁcient 224, 229 disturbances characteristic parameters 24, 49, 50, 54, 56, 58, 248 decay 23, 24, 44–46, 53, 244, 248 fundamental disturbances (see also: ﬂow structure) 224, 226, 231 pulsation 109, 119, 120, 219 drag coefﬁcient 204 of cylinder 202 drag force 203, 205 drag force ﬂow meter 201, 208 eddy viscosity 6, 26, 32 error shift 229, 231, 239, 245, 247, 250

ﬂow angle 241 ﬂow conditioner 61, 65, 69, 223 perforated plates 63 tube bundle 62 ﬂow structure (see also: fundamental disturbances) ring vortex 41, 53 single longitudinal vortex 41, 53 source-sink pair 41, 53 source-sink quadrupole 41, 53 vortex pair 41, 53 vortex quadrupole 41, 53 friction factor 8, 24, 27, 28, 31, 43, 52 friction law 8, 19, 31 Hilbert-transform 83, 86 hot-wire anemometer 65 installation 69, 223, 227, 234, 239, 247 bend 56, 63, 90, 108, 117, 208, 217, 219, 241, 244 double bend 70, 118, 241, 244 elbow see bend gate 229 Kármán constant 5 Kármán vortex street Kalman ﬁlter 89

97, 211

lobed-impeller ﬂow meter

109, 122

mean velocity see pipe ﬂow modal value 117, 130, 144

256

Index

Navier–Stokes equations numerical solution 212 numerical simulation grid structure 189, 215 low Mach number 213 solvers 188 three-dimensional 188 Nyquist theorem 81 oriﬁce meter

223, 224, 229

passage contraction 153, 161 pipe ﬂow axisymmetric ﬂow 42, 43 bulk velocity see mean velocity core region 5, 25, 29 critical zone 18 fully developed 1, 26, 29, 52, 61, 204, 224 fully rough 13 inlet ﬂow 24 law of the wall 4 logarithmic law 5 low Reynolds number 18 mean velocity 26, 61, 80, 129, 231 overlap layer 4, 6, 25 shape factor 204 turbulent 1, 23, 137, 138, 204 two-layer structure 4, 24 velocity distribution 10, 12, 29, 32, 33, 117, 125, 137 wall layer 4, 25 PIV 63 light sheet 64 pressure probe 241 probability density function 137, 138, 233 Reynolds number 8, 52, 62 Reynolds stress 73, 74, 76 secondary ﬂow 71, 72, 225 skin-friction velocity 26 slender-channel theory 27 Strouhal number 218 sublayer fence 227 Superpipe Experiment 8, 31, 52 swirl 43, 70, 71, 149, 165, 167, 176, 179, 181, 245 swirl angle 43, 71, 72, 245, 248

swirl number 24, 43, 49, 53 system-theoretical model 133 tenth-value length 37, 41, 53 tomographic reconstruction 125 turbine ﬂow meter 150, 165, 166, 168, 185, 186 accuracy 187 ﬂow ﬁeld 190 pressure shift 196 two-stage 187 turbulence model 6, 33, 214 turbulent ﬂuctuations 68, 73 ultrasonic ﬂow meter 239, 242 ultrasound 79, 95, 111, 129 ultrasound cross-correlation ﬂow meter 80, 111, 129 ultrasound signal 100, 111, 122, 132 amplitude-modulated 80, 114, 116, 145 carrier frequency 97 demodulation 81, 86, 88, 91 modulation 79, 96 multipath arrangement 124 phase-modulated 80, 88, 89, 114, 116, 145 undersampling 81 velocity defect law 6 Venturi 224, 229, 241 volumetric ﬂow rate 226 vortex ﬂow meter 92, 95, 211, 216 vortex shedding see bluff body vortex street see Kármán vortex street vortical structure 130 wall roughness equivalent sandgrain roughness 14 fully rough 14 hydraulically smooth 14 Kármán’s law 13 natural roughness 14 Nikuradse roughness 13 roughness function 16 roughness parameter 13, 16, 17 roughness Reynolds number 13 sandgrain roughness 13 wall shear stress 36, 226, 227 Wiener–Chintschin theorem 121